The IRGC said the attack destroyed what it called Amazon’s central data hub in Bahrain. Its statement presented the operation as retaliation for US military strikes on Iranian targets, including the Darkhovin nuclear power project in Khuzestan province.

Neither Amazon nor Bahraini authorities immediately confirmed that a cruise missile had directly hit an AWS site during the operation described by Tehran. The claim emerged while Bahrain and other Gulf states were responding to missile and drone threats linked to the widening confrontation between Iran and the United States.

AWS acknowledged that its Middle East region in Bahrain remained unavailable because of damage caused by the conflict. Its service dashboard advised affected customers to restore workloads in other regions, indicating that disruption extended across multiple cloud products and was not expected to be resolved quickly.

The available updates do not establish that the outage was caused by the specific strike announced by the IRGC. AWS infrastructure in Bahrain had already sustained conflict-related damage before Tehran issued its latest statement, complicating efforts to determine whether a fresh attack occurred or whether Iran was referring to an existing operational failure.

Iranian accounts said the missiles also targeted US air-defence equipment and radar installations around Muharraq and Riffa. Bahrain hosts the headquarters of the US Navy’s Fifth Fleet and has become an important logistical and communications centre for American military operations across the Arabian Gulf, Red Sea and north-western Indian Ocean.

Amazon opened the AWS Middle East region in Bahrain in 2019, its first cloud region in the Middle East. The system was designed around three separate availability zones, allowing organisations to distribute applications and data across physically distinct locations while remaining within the same national jurisdiction.

The Bahrain region serves government agencies, banks, telecommunications operators, start-ups and large companies seeking lower data-transfer delays and compliance with local data-residency requirements. Disruption can therefore affect far more than Amazon’s internal operations, potentially interrupting payment services, business software, mobile applications and digital government platforms.

Cloud providers ordinarily use redundancy to keep services operating when individual servers or buildings fail. Prolonged regional damage, however, can force customers to shift applications to facilities elsewhere, a process that may require additional technical work, regulatory approval and higher network costs. Organisations that stored essential systems in only one region face greater exposure.

The attack claim marks a further expansion of the types of infrastructure being identified as military targets. Data centres have traditionally been treated as commercial assets, but their role in government administration, artificial intelligence, intelligence analysis and military logistics has made them strategically significant during armed conflict.

Tehran has increasingly portrayed US technology companies as part of Washington’s security architecture. That approach raises questions about the protection afforded to privately operated digital infrastructure, particularly when the same cloud platforms support civilian businesses and public services alongside government customers.

Damage to cloud facilities also presents risks that differ from conventional attacks on energy or transport infrastructure. Service failures can spread across borders within seconds, while the physical location of affected customer data may be unclear to users. A disabled facility can disrupt companies operating hundreds or thousands of kilometres from the strike site.

Bahrain has strengthened civil-defence measures as hostilities have moved closer to Gulf population centres. Warning systems have been activated during suspected Iranian attacks, and authorities have urged residents to follow official instructions and avoid circulating unverified information.

The confrontation has already placed pressure on regional aviation, shipping, energy markets and telecommunications networks. Missile and drone activity has been reported across Bahrain, Kuwait and Jordan, while US forces have continued strikes against Iranian missile, drone, surveillance and air-defence sites.

For Amazon, the Bahrain disruption adds to the challenge of maintaining cloud availability during a conflict affecting several technology hubs. The company has directed customers towards recovery arrangements outside damaged regions, underlining the importance of geographically separated backups and tested continuity plans.

The article Iran claims strike on AWS hub in Bahrain appeared first on Arabian Post.

The two decisions, adopted at the commission’s July 22 meeting, combine a tougher national security approach towards Chinese technology with an effort to accelerate 5G capacity and prepare networks for future 6G services. They could reshape electronics supply chains, mobile investment plans and competition for valuable mid-band airwaves.

The new equipment rule closes a regulatory gap that allowed products made by other companies to reach US shelves even when they contained chips or communications components supplied by businesses on the FCC’s Covered List. Huawei and ZTE are among the companies placed on that list because Washington considers their equipment a national security risk.

Since 2022, new models produced directly by covered manufacturers have been unable to obtain FCC authorisation. Products incorporating their components, however, could still qualify if the finished device carried another company’s brand. The revised policy prohibits authorisation of equipment containing Huawei-made logic-bearing hardware, extending scrutiny from the identity of the final manufacturer to the origin of critical internal parts.

Logic-bearing hardware can include semiconductors and communications components that process information or control how a device operates. Security officials have argued that compromised parts may create vulnerabilities across an entire product, even when assembly and branding are handled by an unrelated manufacturer.

FCC Chairman Brendan Carr said the commission was fully closing the component-part loophole. Huawei had not issued a public response immediately after the vote.

The change is likely to increase compliance demands for electronics manufacturers, importers and testing laboratories. Companies seeking permission to market wireless products in the US may need stronger documentation showing where chips, modules and other sensitive components were designed and manufactured.

Supply-chain tracing could become particularly important for routers, surveillance systems, connected appliances and industrial communications equipment. Smaller manufacturers that rely on third-party circuit boards or modules may face added costs when identifying the origin of embedded components.

The action forms part of a wider campaign against communications equipment linked to covered Chinese companies. The FCC has expanded restrictions on previously authorised products, foreign-made routers and certain drone systems. It is also examining whether telecommunications carriers should be prevented from interconnecting with covered companies operating data centres or internet exchange facilities in the US.

China has repeatedly rejected US national security allegations against its technology companies and has accused Washington of using security concerns to restrict commercial competition. American officials maintain that communications infrastructure and connected devices require stricter safeguards because vulnerabilities can affect government networks, businesses and consumers.

Alongside the equipment restrictions, the FCC approved rules for auctioning 160 megahertz between 3.98 and 4.14 gigahertz. Combined with the lower C-band already assigned for mobile use, the allocation creates a contiguous 440-megahertz block stretching from 3.70 to 4.14 gigahertz.

Mid-band spectrum is prized by mobile operators because it provides a balance between coverage and capacity. Lower frequencies can travel farther but generally offer less bandwidth, while millimetre-wave signals provide high speeds across shorter distances. C-band frequencies can support wider coverage while carrying large volumes of data.

The commission plans to complete the upper C-band auction by July 2027. The sale will offer 3,248 flexible-use licences across the contiguous US and is expected to generate tens of billions of dollars for the Treasury. The lower C-band auction completed in 2021 raised about $81 billion in gross proceeds.

Verizon, AT&T and T-Mobile secured much of the spectrum offered in the earlier sale. The next auction could attract those operators as they seek additional network capacity, although satellite companies, new wireless entrants and technology groups will also watch the licensing structure closely.

Commercial operations in the upper C-band will not begin immediately. Satellite providers using the frequencies must first clear or reorganise their services. Wireless deployment in the 75 largest markets is scheduled to become possible from December 31, 2030, with remaining markets following from July 1, 2031.

The framework identifies Empresa, Eutelsat, Hispasat, SES and Telesat as incumbent satellite operators. Winning bidders will reimburse eligible relocation costs, while incentive payments will be available to operators meeting clearing deadlines.

The article FCC tightens Huawei curbs and expands 5G spectrum appeared first on Arabian Post.

The Galaxy Z Fold8, Galaxy Z Fold8 Ultra and Galaxy Z Flip8 were introduced at the company’s Galaxy Unpacked event in London on Wednesday. Samsung is expanding its foldable portfolio at a time when the wider smartphone market is shrinking and competition in premium devices is set to intensify.

The standard Fold8 adopts a shorter and wider design intended to make video viewing, gaming and multitasking more comfortable. Its proportions resemble a passport when closed, marking a shift from the taller book-style shape used in earlier generations.

Samsung has priced the Fold8 at $1,899 in the United States. The more powerful Fold8 Ultra will cost $2,099, while the compact Flip8 carries a $1,199 price tag. The Ultra and Flip models are each $100 more expensive than their predecessors, reflecting rising production expenses and the growing cost of advanced components.

The company is seeking to protect its lead in foldable phones before Apple enters the category. Apple is widely expected to introduce its first foldable iPhone later this year, potentially bringing a larger group of premium users into a market that Samsung helped establish.

Foldable devices still account for less than 2 per cent of global smartphone sales, but shipments are forecast to grow by about 20 per cent this year. That contrasts with an anticipated double-digit contraction in the overall handset market, where demand is being weakened by higher prices, economic uncertainty and longer replacement cycles.

Samsung’s launch strategy suggests that it expects foldables to remain one of the industry’s few areas of meaningful growth. The addition of a third model gives buyers a broader choice between portability, screen size and productivity features.

The Fold8 Ultra is aimed at customers seeking a tablet-like experience. It offers the largest display in the range, a high-resolution 200-megapixel camera and a 5,000mAh battery. The standard Fold8 has a smaller body and is positioned as a more portable option for users who still want a large internal screen.

The Flip8 retains the clamshell design that folds vertically into a compact square. Its external FlexWindow has been updated to show more notifications, controls and artificial intelligence features without requiring users to open the device.

All three phones include Samsung’s Galaxy AI tools and Google’s Gemini assistant. The software is designed to summarise content, translate conversations, edit images and provide contextual suggestions across applications. Samsung is placing greater emphasis on AI as hardware improvements become less dramatic and consumers demand stronger reasons to upgrade.

The larger screens on the Fold8 and Fold8 Ultra allow several applications to run at once. Samsung has also introduced software that can suggest actions based on what appears on the display, such as creating calendar entries, comparing information or moving content between apps.

Qualcomm’s latest premium mobile processor powers much of the range, while some versions of the Flip8 will use Samsung’s Exynos chips depending on the market. The mix enables Samsung to reduce reliance on outside suppliers while promoting its own semiconductor technology.

However, rising memory prices pose a complex challenge for the company. Samsung benefits from stronger chip prices through its semiconductor division, but the same trend increases costs for its smartphone business. Memory components form a substantial part of the bill of materials for premium handsets, particularly models offering high storage capacity and advanced AI functions.

The mobile division faces pressure to maintain margins without pushing foldable prices beyond the reach of consumers. Higher retail prices may restrict demand, particularly when conventional flagship phones offer many of the same cameras, processors and AI services at lower cost.

Analysts have also questioned whether a wider display and thinner body will be enough to persuade users to replace existing foldables. Durability, battery life and the visible crease in folding screens remain key concerns, although Samsung says it has strengthened the hinge and improved the internal display structure.

The Fold8 is lighter and slimmer than previous book-style models, addressing complaints that foldables can feel bulky when closed. Samsung has also worked to reduce the prominence of the crease across the main screen and improve one-handed use.

The article Samsung widens foldable range as component costs climb appeared first on Arabian Post.

The service is available through the Gemini Enterprise Agent Platform and can also operate as a component of Google AI Threat Defense. It marks the transition of CodeMender from a Google DeepMind research project into an enterprise product that can be integrated with development tools and software delivery pipelines.

CodeMender differs from conventional code scanners by attempting to establish whether a detected weakness can actually be exploited. The agent creates proof-of-concept exploit code and runs it inside an isolated sandbox managed by the customer. This allows security teams to distinguish practical threats from theoretical findings and direct resources towards flaws presenting the greatest risk.

After confirming exploitability, the agent produces a patch, tests the change and supplies it to developers as a code difference for review. CodeMender does not automatically commit changes to repositories or push them into production. A developer must examine and approve each proposed fix, maintaining a human checkpoint within the remediation process.

Google said the agent can scan for memory corruption, injection vulnerabilities, web security defects, cryptographic weaknesses and unsafe data handling. It supports C and C++, Go, Java, Python, Ruby, Rust and TypeScript, covering languages widely used in enterprise applications, cloud services and open-source infrastructure.

The underlying system combines artificial intelligence models with security tools such as static analysis, dynamic analysis, fuzzing, differential testing and mathematical solvers. Rather than relying only on patterns associated with known bugs, the agent analyses control flow, data flow, application behaviour and the wider context of a software repository.

Code compilation, testing and exploit execution take place inside customer-controlled local sandboxes or isolated virtual machines. The agent’s reasoning and orchestration functions are hosted through Google’s platform. Organisations can route traffic through their virtual private cloud, while Google says source-code data is isolated, encrypted and subject to a zero-retention policy.

Developers can operate CodeMender through a lightweight command-line interface or integrations with tools including Visual Studio Code. It can also run as a headless agent inside continuous integration and continuous delivery pipelines, allowing security checks and patch generation to become part of routine software development.

The launch comes as security teams struggle with large volumes of scanner alerts, many of which require manual investigation before their significance can be established. False positives can delay remediation by forcing engineers to examine warnings that do not represent a workable attack path. Exploit-based verification is intended to reduce that burden, although running automatically generated attack code introduces operational risks that make strong isolation essential.

Security research has shown that coding agents and their sandboxes can face prompt-injection attacks, supply-chain manipulation and attempts to cross execution boundaries. Malicious instructions concealed in repositories, documentation or dependencies may influence an agent’s actions. Customer-managed environments therefore give organisations more control, but they also place responsibility on administrators to configure, patch and monitor those environments securely.

CodeMender began as a DeepMind project announced in October 2025. During its first six months of development, it contributed 72 security fixes to open-source projects, including codebases containing as many as 4.5 million lines. Its early work included tracing complex memory errors and applying broader safeguards intended to prevent entire categories of buffer-related vulnerabilities.

The agent has been tested by companies including Salesforce, Robinhood and Palo Alto Networks. Google is also positioning it alongside Wiz, the cloud security company it agreed to acquire, and Mandiant, its incident response and threat intelligence operation. Planned integrations would use deployment context from the Wiz Security Graph to identify exposed applications and instruct CodeMender to generate targeted repairs.

Customers can select from multiple models depending on the required balance of speed, cost and scanning depth. Support for third-party frontier models is planned later this year, potentially making CodeMender less dependent on a single model family.

The article Google launches CodeMender for automated vulnerability repair appeared first on Arabian Post.

An analysis of HTTP traffic across more than 120 countries found that kickoff times were among the strongest influences on digital behaviour. Matches played between midnight and 8am local time frequently pushed internet activity well above normal levels as supporters stayed awake or rose early to watch. Traffic more than doubled in some markets where overnight usage would ordinarily have been low.

Daytime fixtures produced a different effect. Matches beginning during working hours generally caused only limited movement because much of the audience was already connected. Evening games were more likely to reduce browsing as viewers put aside other devices and concentrated on the action.

The contrast was especially visible when teams in different time zones watched the same fixture. Brazil’s Round of 32 match against Japan began during normal waking hours in South America but landed deep into the night in Japan. Traffic in Japan rose to roughly twice its usual level, while activity in Brazil fell below the established baseline as people moved away from routine browsing.

Bosnia and Herzegovina showed a similar divide across matches. Traffic climbed above twice its normal level during a 2am kickoff, but dropped to about 70 per cent of typical activity when the national side played during the evening.

Cloudflare measured these changes by comparing minute-by-minute traffic with the median recorded during the previous four weeks. The method allowed high-volume and smaller internet markets to be assessed on a comparable basis while reducing distortions caused by ordinary daily fluctuations.

Argentina generated the strongest overall pull on global internet activity. Traffic in the typical country moved about 17 per cent away from normal whenever the defending champions played. France, Spain, Brazil, Portugal, Morocco and Norway also ranked among the teams producing the largest worldwide shifts.

Argentina’s quarter-final against Switzerland on July 11 created a greater global traffic movement than any other match, including the semi-finals and final. The fixture moved internet activity by a factor of about 1.26, ahead of the France-Spain semi-final at approximately 1.21.

The figures measured the scale of change rather than whether traffic rose or fell. A sharp decline caused by viewers abandoning other online activity was therefore treated as equally significant as a surge generated by streaming, messaging or social-media use.

Online behaviour also changed during pauses in play. Most participating countries recorded increases in traffic at half-time as viewers reached for phones, checked scores, browsed social platforms or exchanged messages. Croatia and Bosnia and Herzegovina displayed some of the clearest half-time increases.

The tournament’s mandatory three-minute hydration breaks produced smaller but still measurable spikes. Traffic commonly rose as soon as play stopped and declined again when the match resumed, showing that even brief interruptions were long enough to prompt viewers to switch attention to another screen.

Patterns were not uniform. Forty-four countries, covering 101 national-team match appearances, recorded increases during half-time and hydration breaks. Eight countries followed an almost opposite pattern, with traffic falling during those pauses.

Algeria, Tunisia, Jordan, Egypt and the Democratic Republic of Congo were among the markets where streaming appeared to account for much of the difference. Internet use climbed strongly at kickoff, remained elevated during play and declined at half-time when some viewers closed streams or stepped away.

During Algeria’s match against Austria, multimedia and streaming requests drove much of the traffic increase. Activity dropped during half-time before rising again for the second half. Austria, where streaming represented a smaller share of match-related traffic, recorded greater browsing during the interval.

Hydration breaks had less impact in streaming-heavy markets because viewers were unlikely to close a live feed for a three-minute stoppage. A 15-minute interval gave users more time to disconnect, move away from the screen or switch to another activity.

Traffic to sports-betting websites also rose after the tournament began. Before the opening match, requests followed a clear weekly cycle. Once fixtures started, activity became steadier because matches were played on an almost daily basis, generating more continuous demand.

The article World Cup reshaped global internet traffic patterns appeared first on Arabian Post.

The penalty, announced on July 20, is the largest imposed under the European Union’s Digital Services Act and marks a significant escalation in Brussels’ enforcement campaign against global technology and e-commerce platforms. Regulators said AliExpress had not adequately identified or reduced systemic risks facing consumers despite operating one of Europe’s largest online shopping services.

The Commission found shortcomings in the platform’s handling of counterfeit clothing, unsafe toys, hazardous cosmetics and other products that failed to meet European safety requirements. It also identified weaknesses in systems designed to detect prohibited listings, sanction repeat offenders and prevent removed goods from being offered again.

AliExpress said the fine was disproportionate and indicated that it would appeal. The company maintained that it had cooperated with regulators, strengthened product monitoring and introduced additional safeguards as compliance expectations developed during the investigation.

The Commission opened formal proceedings against AliExpress in March 2024, examining whether the company was meeting obligations applied to very large online platforms. Those services face stricter requirements because their scale can amplify consumer harm and allow unlawful products or content to reach millions of people.

AliExpress had about 193 million monthly users across the European Union during the latest reporting period, making Europe its biggest overseas market. Its designation as a very large online platform requires it to carry out detailed risk assessments, deploy effective mitigation measures and provide regulators with evidence that those systems work.

Investigators concluded that the company’s assessments underestimated the likelihood that consumers would encounter unlawful goods. Enforcement resources were also considered insufficient for the volume of listings, while some sellers were able to evade controls by placing goods in misleading categories or modifying product descriptions.

The Commission said AliExpress had introduced measures including brand-authorisation procedures and improved seller checks, but these steps had not delivered adequate protection. Products reported as illegal could remain available for extended periods, while enforcement against merchants with repeated violations was judged inconsistent.

The company must submit a comprehensive action plan by October 20 explaining how it will correct the deficiencies. Brussels could impose further measures if AliExpress does not provide satisfactory remedies or fails to implement them within the required timetable.

The decision establishes a new financial benchmark for enforcement of the Digital Services Act. Social media platform X was previously fined €120 million over transparency breaches, while Temu received a €200 million penalty connected to illegal-product risks. Shein has also faced scrutiny over its compliance with European digital and consumer-protection rules.

The legislation allows fines of up to 6% of a company’s global annual turnover. Alibaba generates tens of billions of dollars in yearly revenue, meaning the €550 million penalty remains below the maximum available to regulators, although it represents a substantial direct cost and could increase compliance spending across the group.

The action also reflects growing concern about the surge in low-value parcels entering Europe through marketplaces based outside the bloc. Billions of small packages are shipped to European consumers each year, creating difficulties for customs agencies and product-safety authorities seeking to inspect goods individually.

European officials have been considering additional handling charges for low-value parcels and tighter enforcement of customs, tax and safety rules. Such changes could affect the price advantage enjoyed by AliExpress, Temu and Shein, whose business models rely heavily on direct shipments from manufacturers and merchants to consumers.

Consumer groups have repeatedly highlighted risks involving electrical equipment, children’s products, cosmetics and household goods sold through third-party marketplaces. Counterfeit products also create losses for established brands and may expose buyers to materials or components that have not undergone required testing.

For Alibaba, the dispute adds regulatory pressure at a time when it is seeking stronger international growth outside its core domestic operations. AliExpress has expanded delivery networks, marketing campaigns and local merchant programmes across Europe as competition intensifies among discount-shopping platforms.

The article EU imposes record penalty on Alibaba’s AliExpress appeared first on Arabian Post.

Businesses that invested heavily in generative AI are now confronting a widening gap between formal deployment plans and how staff use the technology. Workers are often turning to free consumer platforms or personal accounts because approved systems are slower, restricted or poorly suited to the task at hand.

This informal adoption, widely described as “shadow AI”, can produce productivity gains that remain invisible to management. It can also distort measurements of whether licensed corporate tools are delivering value because employees may switch between approved and unapproved services without recording how either affects output, costs or working time.

Surveys indicate that 86% of employees use AI for work-related tasks at least once a week, while 58% rely on unapproved or publicly available services instead of company-provided platforms. About 63% consider it acceptable to use AI without clearance from information technology departments, and 60% believe meeting a deadline can justify the security risk.

The behaviour is not usually driven by malicious intent. Staff use AI to summarise documents, analyse spreadsheets, rewrite emails, generate computer code and prepare presentations. The danger arises when contracts, customer records, financial projections, employee details, software source code or strategic plans are entered into services that have not been assessed by corporate security teams.

Research into workplace practices has found employees sharing datasets, personnel information and financial or sales material with unauthorised tools. Such disclosures can breach confidentiality obligations, privacy rules and contractual restrictions even when no conventional cyberattack has occurred.

Companies also struggle to detect the activity because AI platforms operate through ordinary web traffic and resemble standard productivity applications. Traditional data-loss prevention systems were designed to identify files being copied, emailed or transferred. They may not recognise sensitive information pasted into a chatbot, converted into prompts or embedded within AI-generated workflows.

The governance problem has intensified as boards demand evidence that spending on models, cloud computing, licences and specialist staff is producing measurable results. Reported returns vary sharply. Some early adopters claim positive and quantifiable gains, while other surveys show only a minority of organisations achieving significant returns from generative AI or autonomous agents.

A study of large listed companies found that deep integration of AI into business processes remained limited despite widespread experimentation. Only 11% of S&P 500 companies had deeply embedded AI in operations by 2025, while another 10% used it directly in producing goods or delivering services. Researchers found no clear difference in capital expenditure or productivity across the wider sample, highlighting the difficulty of separating real transformation from corporate promotion.

The measurement challenge reflects the way AI is being introduced. Many companies count licences, users or chatbot interactions rather than tracking changes in revenue, error rates, processing time, customer retention or operating costs. Time saved by one employee may also be offset by another worker checking inaccurate outputs, correcting fabricated information or reviewing AI-generated code.

Generative systems produce probabilistic answers rather than predictable results, making their value dependent on the task, data quality and level of human supervision. Gains may appear in faster decisions or improved service rather than direct cost reductions, while benefits can take months to emerge after training and workflow redesign.

Security specialists argue that simply banning public tools can push usage further underground. Employees are more likely to comply when approved platforms offer comparable speed and functionality, with straightforward access and clear rules on what information may be submitted.

Companies are responding by creating inventories of AI applications, restricting access to high-risk services and introducing enterprise accounts that prevent customer data from being used to train public models. Some are also installing systems that identify sensitive prompts, redact confidential details and maintain records of employee interactions.

Training is becoming another priority because many workers do not distinguish between personal and corporate accounts or understand how data retention policies differ between services. Clear guidance is particularly important in finance, healthcare, legal services and human resources, where a seemingly routine request for a summary may expose regulated or personally identifiable information.

The article Shadow AI use clouds returns and exposes data appeared first on Arabian Post.

The article London exchange prepares AI-ready overnight trading venue appeared first on Arabian Post.

The article Google broadens Gemini range with lower-cost models appeared first on Arabian Post.

The article SpaceX faces $116 billion insider share release appeared first on Arabian Post.

The S&P 500 gained about 0.6%, while the technology-heavy Nasdaq Composite advanced roughly 0.9%. The Dow Jones Industrial Average posted a smaller rise as traders concentrated on semiconductor shares, corporate earnings and the comforting possibility that another expensive computer might solve everything.

Nvidia climbed around 1.5%, joining Advanced Micro Devices, Intel, Micron Technology, Marvell and other chipmakers in a broad recovery. The Philadelphia Semiconductor Index rose about 2.5% after entering bear-market territory, a technical condition that occurs when investors suddenly remember that share prices can move in both directions.

Market strategists described the rebound as evidence of resilience, renewed confidence and the enduring human ability to forget Friday by Monday morning.

“Investors have carefully examined last week’s losses and concluded that they did not enjoy them,” said one fictional Wall Street strategist, adjusting a chart until the line pointed upwards. “The fundamentals remain strong, especially the fundamental belief that someone else will buy at a higher price.”

The rally followed a sharp retreat in technology shares that had tested enthusiasm for the artificial-intelligence boom. Semiconductor stocks have powered much of the market’s advance, supported by huge spending on data centres, processors and cloud infrastructure. Their valuations have also required investors to imagine several prosperous decades in advance while politely ignoring the electricity bill.

Alphabet rose more than 3% after reports of new artificial-intelligence chips, giving traders another opportunity to celebrate a company spending billions to compete against other companies spending billions. The apparent objective is to build machines intelligent enough to explain why all the spending was necessary.

AMD advanced after expanding its artificial-intelligence partnership with Microsoft, reinforcing the industry’s preferred business model in which one technology giant sells equipment to another technology giant so both can announce accelerating demand.

Sandisk, Micron, Intel and Marvell also gained, helping the semiconductor sector recover part of its previous losses. The iShares Semiconductor exchange-traded fund rose more than 1%, confirming that diversified exposure can enable investors to worry about several chipmakers simultaneously.

Attention now turns to quarterly results from Alphabet, Tesla, Intel, IBM and Texas Instruments. Analysts expect the announcements to determine whether soaring technology valuations are supported by earnings, revenue, guidance or sufficiently confident use of the phrase “long-term opportunity”.

Companies failing to beat forecasts, raise forecasts and express excitement about artificial intelligence could face punishment. Merely producing record profits may no longer be enough, since Wall Street increasingly requires corporations to outperform expectations that were raised because they previously outperformed expectations.

Second-quarter earnings for companies in the S&P 500 are projected to rise strongly, with technology firms expected to provide a large share of the growth. The market’s tolerance for disappointment, however, has narrowed as valuations have expanded and investors have become accustomed to results that would once have been considered extraordinary.

Artificial-intelligence expenditure by major cloud providers remains a central pillar of the rally. Data centres are consuming more advanced processors, memory chips, networking equipment and power, creating a vast industrial ecosystem dedicated to helping office workers produce meeting summaries they will not read.

Executives insist the investments will improve productivity, transform industries and unlock new revenue streams. Investors have broadly accepted those assurances while requesting quarterly proof, daily share-price appreciation and immediate confirmation that spending with no obvious limit will generate returns on a convenient timetable.

Outside technology, Domino’s Pizza gained after reporting stronger revenue, demonstrating that markets still recognise older forms of innovation such as placing cheese on bread and delivering it to people who have spent the day refreshing their brokerage accounts.

AMC Entertainment also surged after better-than-expected revenue, reminding traders that cinemas remain attractive whenever audiences are offered films, air conditioning and temporary shelter from financial television.

The gains came despite elevated Treasury yields, geopolitical tension and concerns about energy costs. The yield on the 10-year Treasury moved towards 4.6%, presenting investors with the unfamiliar possibility of earning a return without studying semiconductor supply chains or pretending to understand neural-network architecture.

Oil prices remained volatile as disruptions around the Strait of Hormuz sustained inflation fears. Market participants acknowledged the risks before returning to the more urgent question of whether Nvidia might rise another 2% before lunch.

Trading desks said the chip recovery had restored order after last week’s slump, although caution remained visible in the widespread practice of describing every purchase as “selective”.

By the closing hours, Wall Street had once again established that artificial intelligence was either a historic transformation, an expensive experiment or both, depending on the next earnings slide.

The article Chip rebound restores Wall Street’s artificial confidence appeared first on Arabian Post.

The editor supports headings, tables, numbered and bulleted lists, quotations, code blocks, mathematical formulas and media embedded within the body of a message. Users can compose posts of more than 32,000 characters, removing the need to divide detailed announcements, reports or explainers across multiple messages or link readers to an external publishing platform.

The tool is initially available to Telegram Premium subscribers and has been designed for use on both mobile devices and desktop computers. Its visual interface allows people to format material without relying on Markdown, HTML or other specialist syntax, although Telegram’s bot and developer systems continue to support structured formatting through code.

Images, videos and other media can be positioned alongside text rather than appearing only as separate attachments below a message. This gives channel administrators, businesses, educators and independent publishers greater control over how information is presented and read inside the application.

Telegram has also integrated artificial intelligence tools that can generate or rewrite text, build tables and create formulas. The company describes the system as privacy-conscious, positioning it as an assistant for preparing material rather than a replacement for the person publishing it. Users remain responsible for reviewing generated output before distribution, particularly when messages contain figures, technical information or factual claims.

The feature places Telegram closer to blogging services, collaborative document tools and newsletter platforms while retaining the distribution advantages of a messaging application. Channel owners can publish formatted articles to subscribers without directing them to a website, while organisations can use the editor for product notes, policy updates, schedules, research summaries and internal documentation.

Telegram already operated Telegraph, a lightweight web publishing service used to create standalone articles that could be opened through links. The new editor brings many of those publishing functions into conversations themselves, reducing the number of steps required to create and distribute long-form material.

The shift may be particularly significant for publishers and creators whose audiences primarily consume information through smartphones. Tables, clear headings and embedded illustrations can make complicated subjects easier to follow on smaller screens, although excessively formatted messages may also feel cumbersome in fast-moving group conversations.

Long posts will remain subject to the same forwarding, sharing and moderation systems that apply to other Telegram messages. Channel administrators can edit published material, while readers can react, reply or forward it depending on the permissions selected for the chat.

The editor forms part of a wider Telegram update that also introduces Communities, a system for organising related groups and channels under a common structure. Communities can bring announcements, discussions and specialised groups together, helping large organisations manage audiences that previously had to navigate several unconnected chat links.

Telegram has additionally expanded the way bots operate inside conversations. Ephemeral bot messages can appear temporarily and disappear after completing their purpose, potentially reducing clutter created by automated confirmations, menu prompts and service notifications.

Guest bots can be mentioned inside private or group chats even when they have not been formally added as members. This allows users to call an artificial intelligence assistant, information service or automated tool for a specific task without granting it permanent participation in the conversation.

Such access could make fact-checking, translation, image generation and task automation easier, but it may also raise questions over what information is shared with third-party bots. The privacy and reliability of a bot can depend on its developer, making careful selection important when conversations contain confidential or commercially sensitive material.

Telegram has also broadened GIF search results, giving users more options when searching for animated responses. The change is comparatively small but reflects the company’s effort to improve both long-form publishing and everyday communication within the same release.

The rich text editor strengthens Telegram’s position as a hybrid platform combining messaging, broadcasting, automation and content publishing. Rival services offer selected elements of this model, including community management, channels, artificial intelligence assistance and formatted business messages, but few allow users to send document-like posts as ordinary messages.

The update could alter how companies and media organisations use Telegram channels, especially in markets where the application already functions as a major distribution network. Publishers may be able to deliver visually structured reports without maintaining separate mobile pages, while businesses could issue manuals, catalogues and detailed customer notices inside existing chats.

The article Telegram launches rich editor for long-form messages appeared first on Arabian Post.

The entrepreneur and strategist argues that the debate around AI and creativity has been captured by the wrong question

TORTOLA, British Virgin Islands —

The conversation about artificial intelligence and creative work has been dominated by a single, persistent question: will AI replace human creativity? Alessio Vinassa thinks this is the wrong question, asked at exactly the wrong time, by people who fundamentally misunderstand what creativity is.

“Creativity is not a skill set,” Vinassa said. “It is a form of output that emerges from a particular kind of attention — the kind that asks why something matters, not just how it works. That form of attention is not a technical process. AI does not have it. What AI has is something complementary and powerful: the capacity to execute, at extraordinary speed and quality, the vision that human attention produces.”

Vinassa, who is actively developing ventures at the intersection of AI and content creation, draws an analogy to the introduction of photography in the nineteenth century. “When the camera arrived, people asked whether it would end painting,” he said. “It didn’t. What it did was liberate painting from the obligation to be documentary — which is how you get Impressionism, Expressionism, and every significant art movement of the modern era. The camera changed what painting needed to be. AI will do the same for everything it touches.”

His view is informed by his recently published book No One Is Coming: The Mental Operating System for Leaders Under Pressure, which examines how people process complexity, make decisions under uncertainty, and build internal capacity for performance. Vinassa applies the same framework to creative and organizational questions: the technology is only as valuable as the human operating system behind it.

“The leaders and creators who will build the most significant things with AI are not the technically proficient ones,” he said. “They are the ones who have something to say — who have done enough internal work, lived enough real experience, and developed enough genuine perspective that the tools have material to work with. Technical fluency in AI without vision behind it produces content that everyone can see through immediately.”

This point extends to the economics of creative production. Vinassa argues that AI has dramatically lowered the cost of the execution layer — the translation of a creative concept into a finished artifact — while simultaneously raising the premium on the conceptual layer: the capacity for original thought, emotional intelligence, and genuine narrative authority that no technology can replicate. “If you are a creator who was competing primarily on technical execution — on the ability to render, to code, to produce — AI has disrupted your position,” he said. “If you are competing on the quality of your ideas and the depth of your perspective, AI has given you a significant advantage.”

The practical implication, for Vinassa, is that the current moment rewards investment in what he calls “creative infrastructure”: the development of perspective, voice, and original thinking that makes any tool powerful in the hands of the person using it. “The most valuable thing a creator can build right now is not a technical capability,” he said. “It is a point of view that no algorithm has yet produced. That remains, for the foreseeable future, irreplaceable.”

Vinassa is currently engaged in multiple ventures exploring the use of AI in narrative production, digital identity, and creative content at scale. He describes the current moment as “genuinely historic — not because the tools are impressive, but because the combination of powerful generation capability and genuine human authorship is producing things that were categorically impossible three years ago.”

ABOUT ALESSIO VINASSA

Alessio Vinassa is a serial entrepreneur, business strategist, and thought leader focused on leadership, adaptability, and sustainable growth in global markets. His work spans technology, AI, venture building, and human performance, mentoring founders and executives as they navigate complexity, build resilient organizations, and align long-term strategy with execution discipline. For more information, visit www.alessiovinassa.io.

The article Alessio Vinassa: ‘Generative AI Is the Most Important Creative Tool Since the Camera — and the Most Misunderstood’ appeared first on Arabian Post.

The company said Fable 5 would remain available on Pro, Max, Team and eligible Enterprise subscriptions for up to half of each user’s weekly allowance. Claude Code customers will also retain weekly usage limits that are 50 per cent higher until the new deadline, set for 11.59pm Pacific Time.

The one-week extension is the third change to the access timetable since Fable 5 returned to Anthropic’s services at the start of July. The initial inclusive period was due to expire on July 7 before being moved to July 12 and then July 19, giving subscribers additional time to test the company’s most capable generally available model without buying separate usage credits.

Anthropic has not indicated whether access will be extended again. Under the planned arrangement, subscribers who want to continue using Fable 5 after exhausting any inclusive allowance will have to pay through usage credits, bringing the consumer experience closer to the token-based billing already applied to developers using the Claude application programming interface.

Fable 5 costs $10 per million input tokens and $50 per million output tokens through the API, twice the corresponding rates for Claude Opus 4.8. It supports a one-million-token context window, output of up to 128,000 tokens and adaptive reasoning designed for long-running coding, research and enterprise tasks.

The higher cost reflects the computing resources needed to operate frontier models over large codebases and extended autonomous sessions. It also illustrates the financial challenge facing artificial intelligence companies as subscription customers increasingly use models for workloads that can consume millions of tokens during a single project.

The extension may help Anthropic measure demand and ease subscriber concerns before separating access to Fable 5 from fixed-price plans. Users accustomed to unlimited-looking interfaces have criticised additional metering across the AI industry, particularly when agentic coding tools can generate charges without the predictable interaction pattern of conventional chatbots.

Fable 5 was launched on June 9 alongside Claude Mythos 5, a closely related model with fewer restrictions in selected cybersecurity and scientific areas. Mythos access remains limited to approved organisations, while Fable includes safeguards that may redirect flagged biology or cybersecurity requests to Opus 4.8.

Anthropic temporarily suspended both models in June after US export controls restricted their availability to foreign nationals. Fable 5 was restored globally on July 1 after those restrictions were lifted, beginning the sequence of promotional access windows now extended through July 19.

Attention has also shifted to an unidentified model labelled “Claude Honeycomb EAP” that briefly appeared as an option for some Cursor users. EAP commonly refers to an early-access programme, but neither Anthropic nor Cursor has publicly identified Honeycomb as a forthcoming commercial model.

Online claims linking Honeycomb to a supposed Claude Opus 5 release remain unverified. Anthropic has announced no Opus 5 model, release date or pricing, and the company’s official product catalogue continues to list Opus 4.8 as the current model in the Opus family.

Cursor forum users reported seeing Honeycomb with a large context capacity and an extra-high reasoning setting before the listing disappeared. Temporary model names are often used for internal testing, staged deployments and evaluation builds, meaning the appearance alone does not establish that a public launch is imminent.

The episode nevertheless highlights the role of coding platforms in revealing changes within the competitive AI market. Cursor integrates models from several providers and regularly tests unreleased configurations, creating opportunities for model identifiers or experimental settings to become visible before formal announcements.

Anthropic is competing with OpenAI, Google, xAI and specialist coding providers for developers building software through autonomous agents. Fable 5 has been positioned for complex migrations, multi-stage engineering projects and professional work that requires prolonged reasoning with limited supervision.

The company says the model can operate across large repositories, generate and run tests, inspect visual outputs and revise its own work. Its safeguards, mandatory 30-day data retention and higher token charges may limit adoption among organisations with strict privacy requirements or workloads where lower-cost models provide adequate performance.

The article Anthropic extends Fable access as model rumours intensify appeared first on Arabian Post.

The companies announced a definitive agreement on July 8, with completion subject to regulatory approvals and customary closing conditions. Financial terms were not disclosed. Kentik will continue operating under its existing contracts, pricing, account teams and support arrangements while the transaction moves towards completion.

The deal combines Infoblox’s strengths in Domain Name System services, Dynamic Host Configuration Protocol, IP address management and asset discovery with Kentik’s platform for monitoring traffic flows, application performance, routing paths, cloud networks and digital user experiences.

Infoblox said the acquisition would expand its visibility beyond network identity and configuration data into continuously updated intelligence showing how traffic is moving, where performance is changing and which infrastructure problems require immediate attention.

The combination is intended to create a unified operational data layer across on-premises systems, public clouds, applications and internet infrastructure. That data could be used by network, security and cloud teams, as well as artificial intelligence agents designed to investigate incidents, prioritise problems and recommend remedial action.

Infoblox chief executive Scott Harrell said every device, application and cloud workload running through the company’s technology generated valuable operational context. Kentik would enrich that information with real-time hybrid-cloud intelligence, helping customers improve resilience, performance and security.

Kentik chief executive Avi Freedman said the companies planned to build a common operational foundation for the next generation of networking. Kentik, founded in 2014, has developed its business around converting network telemetry into answers for operators managing distributed and cloud-based infrastructure.

The acquisition arrives as businesses deploy more AI workloads and digital services across environments that may include private data centres, multiple cloud providers, branch locations and third-party networks. Traditional monitoring tools often provide separate dashboards for devices, applications, security alerts and traffic patterns, leaving operators to correlate information manually.

Infoblox and Kentik are seeking to address that fragmentation by combining authoritative network records with live telemetry. DNS data can reveal which services, users and devices are communicating, while flow records and routing information can show how traffic moves and where disruption or congestion has emerged.

The deal also strengthens Infoblox’s push into so-called agentic operations, where AI systems perform multi-step operational tasks rather than merely generating summaries. Reliable infrastructure data is considered essential for such systems because incomplete inventories or outdated configuration records can lead automated tools to misdiagnose incidents or recommend unsafe changes.

Infoblox launched Infoblox IQ in June as an agentic operations layer for networking and security. The system analyses DNS queries, DHCP leases, IP assignments, device activity and security events, allowing teams to investigate conditions through natural-language commands and automate parts of incident response.

The company has said one deployment reduced more than 504,000 operational events to 24 prioritised actions. Investigations that previously took between 45 and 90 minutes were surfaced immediately with supporting context. Infoblox IQ for Threat Defense became available during June, while additional DDI, assistant and Model Context Protocol capabilities are scheduled for wider availability later in 2026.

Infoblox also expanded its Universal Asset Insights platform in June, adding more than 40 third-party application programming interface integrations across cloud, networking, endpoint, identity, security and operational technology systems. Those additions are designed to maintain a continuously updated inventory rather than rely on periodic network scans.

Kentik’s technology could broaden that strategy by supplying telemetry from routers, cloud platforms, synthetic tests, internet routes and application traffic. The company’s platform is used to identify outages, detect abnormal traffic, analyse capacity, assess service performance and trace problems affecting customers.

The acquisition reflects a wider convergence between network management, cybersecurity and observability. As infrastructure becomes more distributed, security incidents can resemble performance failures, while routing changes, cloud misconfigurations and capacity problems can expose services or disrupt business operations.

Network observability providers are therefore moving beyond dashboards towards automated diagnosis and predictive analysis. At the same time, DNS and IP address management vendors are adding asset intelligence, security analytics and cloud automation to defend their role as foundational infrastructure platforms.

The article Infoblox strikes Kentik deal to deepen network intelligence appeared first on Arabian Post.

The lawsuit, filed on Friday in the US District Court for the Northern District of California, marks a sharp escalation in tensions between companies that began working together two years ago. Apple alleges that OpenAI systematically obtained confidential product, manufacturing and supplier information by recruiting staff who had worked on some of its most closely guarded technologies.

OpenAI rejected the central accusation. The company said it had no interest in the trade secrets of other businesses and remained focused on developing technology intended to benefit users. The defendants have not yet filed their formal responses to Apple’s complaint, and the allegations have not been tested in court.

The case names Chang Liu, a former senior systems electrical engineer at Apple, and Tang Yew Tan, a former vice-president of product design who spent 24 years at the iPhone maker. Tan, who worked on products including the iPhone, Apple Watch and iPod, is now OpenAI’s chief hardware officer.

Apple claims Liu retained a company-issued laptop after leaving and exploited an authentication vulnerability to enter its internal network. He allegedly downloaded dozens of confidential files connected to hardware development while preparing to join OpenAI earlier this year.

Tan is accused of sending himself confidential information about suppliers and internal industry assessments before his departure. Apple also alleges that he encouraged job candidates who were still employed by the company to bring physical components to OpenAI interviews for demonstrations and technical discussions.

The complaint also names OpenAI Foundation, its commercial arm OpenAI Group PBC and io Products, the hardware venture associated with former Apple design chief Jony Ive. OpenAI acquired io Products last year in a transaction valued at nearly $6.5 billion, placing Ive and several former Apple designers at the centre of its effort to create a new category of artificial intelligence device.

OpenAI has disclosed few details about the planned product. Executives have described the project as an attempt to move beyond established screens, applications and interfaces, while analysts expect it to provide users with a direct way to interact with ChatGPT. The company’s chief financial officer, Sarah Friar, said in April that consumer hardware was expected towards the end of this year.

Apple argues that the hardware programme rests partly on information taken from its own development operations. It is seeking financial damages and an injunction preventing OpenAI and the other defendants from possessing, disclosing or using the disputed material.

The iPhone maker said it contacted OpenAI in February after identifying signs that confidential information might have reached the artificial intelligence company. Apple claims the approach received no response, prompting it to broaden its internal investigation and uncover what it describes as a pattern of theft involving former employees.

More than 400 former Apple workers are now employed by OpenAI, the lawsuit says. Hiring employees from a competitor is generally lawful in California, where restrictions on workers moving between companies are limited. Apple’s case is therefore likely to depend on whether it can prove that identifiable trade secrets were taken, transferred or used, rather than relying only on the movement of personnel.

The dispute threatens a partnership unveiled in 2024, when Apple integrated ChatGPT into its operating systems. The arrangement allows Siri to direct some user questions to OpenAI’s chatbot and gives device owners access to ChatGPT without leaving Apple’s software environment.

That relationship has become more complicated as OpenAI develops hardware that could compete for consumer attention with the iPhone. Apple has also expanded its work with other artificial intelligence developers, including Google, as it rebuilds Siri and attempts to close the gap with rivals in generative AI.

The lawsuit adds another legal obstacle to OpenAI’s hardware strategy. The company and Ive have also faced trademark and trade-secret litigation from technology start-up iyO, which has challenged the name of the io venture and alleged misuse of confidential product information.

The article Apple targets OpenAI over alleged secrets theft appeared first on Arabian Post.

The services were provided through entities in Singapore connected to Alibaba, Baidu and Tencent. Transactions involving the subsidiaries were permitted under existing US rules, even though their parent groups appear on a Department of Defense list of Chinese military companies.

The arrangements have intensified debate over whether restrictions designed around semiconductors and computing infrastructure can adequately control access to powerful AI systems delivered through cloud platforms. Unlike advanced chips, software models can be accessed remotely and distributed through corporate structures spanning several jurisdictions.

OpenAI and Google said their services are governed by safeguards intended to prevent prohibited activity and detect misuse. Both companies maintain restrictions on direct access from China, where many leading US AI products are unavailable. However, multinational groups can operate subsidiaries in supported markets, creating uncertainty over when the identity and ownership of a customer should determine eligibility.

OpenAI has suspended access for some users linked to Alibaba after detecting activity that raised concerns about model distillation. The technique involves generating large volumes of answers from an advanced model and using those outputs to train another system. It can enable developers to reproduce aspects of a rival model’s capabilities without bearing the full cost of its development.

The company has previously accused China-based developers of attempting to extract knowledge from its systems through distillation. US technology executives and officials regard the practice as a growing threat because frontier models require vast investments in computing power, data, engineering and electricity.

Google has also acknowledged that restrictions can be circumvented, although it says it employs monitoring systems, usage policies and technical controls to identify abusive behaviour. Its AI services are delivered through a range of consumer and enterprise products, including the Gemini model family and the Vertex AI cloud platform.

The Pentagon’s Section 1260H list identifies companies that Washington believes contribute to China’s military-civil fusion strategy. The designation is not the same as placement on the Commerce Department’s Entity List and does not automatically prohibit all commercial dealings.

That distinction has become central to the controversy. Alibaba, Baidu and Tencent are among China’s largest technology groups, with extensive interests in cloud computing, digital payments, online platforms, autonomous systems and AI research. Their global operations include separately incorporated companies outside mainland China.

The Defense Department’s June 2026 assessment said Baidu contributed to China’s defence industrial base through affiliations with government technology bodies. Beijing and affected companies have repeatedly rejected US allegations that commercial enterprises are improperly linked to military activity.

Washington has imposed increasingly strict limits on exports of high-performance processors and semiconductor manufacturing equipment to China. Nvidia and other chipmakers must obtain licences before selling certain advanced products, while restrictions also cover equipment used to manufacture cutting-edge chips.

Controls on access to trained AI models remain less comprehensive. Policymakers face difficulties defining which systems should be restricted, measuring their capabilities and preventing customers from reaching them through cloud accounts registered in third countries.

Security advocates argue that access to leading US models can accelerate research by China-based companies, improve military planning tools and support offensive cyber operations. They have called for ownership-based rules that would apply to subsidiaries controlled by organisations from restricted jurisdictions.

Industry representatives warn that broad controls could damage US cloud providers, encourage customers to adopt competing models and speed the development of independent AI ecosystems outside the United States. They also argue that centrally hosted models offer greater visibility than downloadable software because providers can monitor prompts, block accounts and update safeguards.

Anthropic has adopted a stricter position than several competitors. The Claude developer bars companies controlled from China and other unsupported regions from using its services, including through overseas subsidiaries. It says corporate ownership can create security risks regardless of where an entity is registered.

The dispute is unfolding as China considers tighter controls on foreign access to its own advanced models. Chinese systems from DeepSeek, Alibaba, Moonshot AI and Zhipu AI have gained users abroad because of their competitive performance and lower operating costs.

The article US AI access to China groups draws scrutiny appeared first on Arabian Post.

The test linked an ordinary Android smartphone with a satellite messaging device, allowing SMS and SOS communications to travel in both directions through Space42’s next-generation mobile communications satellite. The companies said the connection required no additional hardware on the smartphone.

Commercial introduction of the messaging services is targeted for the end of 2026, subject to regulatory approvals and agreements with mobile network operators. The development could extend basic communications to deserts, mountains, offshore areas and other locations beyond the reach of conventional mobile towers.

Skylo’s non-terrestrial network platform is compliant with standards developed by the 3rd Generation Partnership Project, or 3GPP, which establishes technical specifications used across global mobile networks. This standards-based approach is designed to allow compatible phones and connected devices to move between terrestrial and satellite coverage without relying on proprietary terminals.

The system also enables users to retain their existing mobile operator identity. Messages can pass through the operator’s SIM and authentication framework without requiring changes to its core network infrastructure. That structure is intended to preserve the relationship between telecom companies and their subscribers while adding satellite coverage as an extension of existing services.

Ali Al Hashmi, chief executive of Space Services at Space42, said the test showed that Thuraya-4 could deliver resilient, carrier-grade messaging directly to devices. He said the capability was aimed at consumers as well as governments, enterprises and mobile operators seeking communications in underserved or isolated areas.

Skylo chief executive and co-founder Parth Trivedi said the trial demonstrated that satellite messaging could function on phones already used by subscribers and through existing operator networks. The companies had earlier completed a two-way, real-time voice call over Thuraya-4 without modifying the SIM card or altering operator infrastructure.

The latest test builds on a partnership announced in May to integrate Skylo’s platform with Thuraya-4. The companies are seeking approvals across target markets, with deployment expected to proceed in stages as regulators and mobile operators clear the service.

Space42 and Skylo are also working on location sharing, application data, internet-of-things transmission and automotive connectivity using the same network layer. Those functions could support vehicle tracking, industrial monitoring, emergency response and remote infrastructure management alongside consumer messaging.

Direct-to-device satellite technology has emerged as a major area of competition within the telecom and space sectors. Satellite operators, handset manufacturers and mobile carriers are seeking to eliminate coverage gaps by enabling phones to connect directly with spacecraft when ground networks are unavailable.

Early commercial services have largely focused on emergency alerts and limited text messaging because these applications require less bandwidth than voice calls or broadband internet. Operators are gradually expanding towards data services as satellite capacity, antenna technology and handset chipsets improve.

Thuraya-4 entered global commercial service in November 2025 after being launched aboard a SpaceX Falcon 9 rocket in January that year. The geostationary satellite provides L-band connectivity across more than 100 countries in Europe, Africa, Central Asia and the Middle East.

Built on Airbus’s Eurostar Neo platform, Thuraya-4 carries a 12-metre reflector antenna and uses a software-defined architecture that allows coverage and capacity to be adjusted from orbit. Its system can deliver speeds exceeding one megabit per second and supports satellite communications across land, maritime and aviation markets.

The satellite underpins Space42’s next generation of mobile communications products and carries a 15-year government contract valued at $708 million. It was developed to replace ageing capacity while supporting new direct-to-device and internet-of-things applications.

Space42 was created in 2024 through the merger of satellite operator Yahsat and geospatial intelligence company Bayanat. Its main shareholders include G42, Mubadala and International Holding Company. The Abu Dhabi-listed group combines satellite communications, artificial intelligence and geospatial data services through its Space Services and Smart Solutions businesses.

Skylo, headquartered in Mountain View, California, operates a standards-based satellite connectivity platform that works with satellite operators, telecom carriers and device manufacturers. Its network already supports consumer, enterprise and connected-device services across several continents.

The article Space42 validates satellite messaging on standard phones appeared first on Arabian Post.

The shift marks a notable change for one of the technology industry’s biggest AI spenders. Microsoft has relied heavily on OpenAI’s models to power Copilot features across its software portfolio, while also adding Anthropic technology for selected Microsoft 365 functions. The company is now routing tens of thousands of user prompts each week through its own Microsoft AI, or MAI, model family, a sign that internal systems are being prepared for high-volume enterprise workloads.

The move does not mean Microsoft is abandoning OpenAI. The two companies remain tied through a long-term strategic partnership, cloud infrastructure agreements and deep product integrations. Microsoft has invested billions of dollars in OpenAI since 2019 and remains one of its most important commercial partners. But the operational balance is changing as AI usage spreads from premium demonstrations into everyday office tasks that can generate large and unpredictable compute bills.

Excel and Outlook are central to that calculation. A spreadsheet formula explanation, email summary, meeting follow-up or draft response may not require the most advanced frontier model available. For high-frequency tasks, Microsoft can use smaller or specialised models that cost less to run while reserving more powerful outside models for complex reasoning, coding or creative work. That approach gives the company more control over margins as Copilot moves deeper into paid workplace subscriptions.

Microsoft AI, led by Mustafa Suleyman, unveiled seven in-house models in June covering reasoning, coding, image generation, transcription and speech. The MAI family includes systems aimed at narrower tasks rather than a single all-purpose model designed to outperform every rival. Company executives have presented the effort as part of a broader strategy to build a full AI ecosystem, not merely resell partner technology inside Microsoft products.

Cost is becoming a defining issue for the AI industry. Large language models consume heavy computing resources, with expenses rising as users ask longer questions, upload larger files and use agentic tools that perform multi-step tasks. Enterprise customers often pay through token-based pricing, where input and output length affect charges. That makes profitability difficult to predict when AI features are embedded into widely used software.

Microsoft’s own financial disclosures show the scale of the infrastructure race. The company has continued to lift capital expenditure to support cloud services, AI training and data-centre capacity. Higher spending has supported Azure growth and helped Microsoft sell AI services to large companies, but investors are watching closely for evidence that AI products can produce durable margins rather than simply drive server demand.

The internal model push also gives Microsoft leverage in a market where dependence on one model provider can create pricing and product risks. By using OpenAI, Anthropic, MAI and open-source models across different use cases, Microsoft can match cost, latency, data sensitivity and performance requirements more flexibly. This multi-model approach is becoming common among large enterprises that want to avoid being locked into a single AI supplier.

OpenAI remains central to Microsoft’s premium AI strategy. Its frontier models continue to power advanced Copilot features and remain deeply integrated with Azure OpenAI Service, which gives corporate customers access to OpenAI systems through Microsoft’s cloud. Anthropic also retains a role in parts of the enterprise AI stack, particularly where customers or product teams prefer Claude models for specific language, coding or workflow tasks.

The latest routing changes are therefore more about optimisation than rupture. Microsoft is effectively creating a tiered AI architecture: expensive frontier models for difficult tasks, lower-cost internal models for routine work, and specialised systems for voice, image, coding or transcription. The company’s advantage lies in owning the application layer, the cloud infrastructure and now a growing set of first-party models.

The article Microsoft shifts AI workload to cut model costs appeared first on Arabian Post.

The main obstacle is no longer only model performance. Deployment is being slowed by multi-factor authentication prompts, short-lived session tokens, bot-detection systems, consent screens, delegated permissions and audit requirements built for people using keyboards, not software agents acting across several applications. The gap has become more visible as companies try to turn generative AI from workplace assistants into task-performing agents that can open tickets, update records, query databases, send emails and complete transactions.

The shift is sharp. Fewer than 5 per cent of enterprise applications had embedded task-specific AI agents in 2025, while that share is expected to rise to as much as 40 per cent by the end of 2026. At the same time, more than 40 per cent of agentic AI projects are expected to be cancelled by the end of 2027 because of high costs, weak business cases or inadequate risk controls. The figures point to a market expanding quickly, but not yet supported by mature operating rules.

The friction often appears at the first practical step: access. An agent may be able to summarise a contract or plan a workflow, but it can fail when asked to log into a customer relationship management system, renew a session, pass MFA, or use a web interface protected by anti-bot controls. Many businesses have tried to work around the problem by sharing service accounts, storing user tokens, or allowing agents to operate under broad permissions. Those shortcuts create security risks because they blur responsibility and make it harder to identify whether an action was taken by an employee, an approved agent, a shadow agent, or a compromised credential.

Identity providers are now racing to define AI agents as first-class non-human identities. Microsoft’s Entra Agent ID framework allows agents to request access tokens and use them to reach services including Microsoft Graph, internal systems and third-party applications. The platform is designed to register agent identities, manage authorisation and apply governance through standard protocols such as OAuth 2.0, Model Context Protocol and Agent-to-Agent communication. Okta has also moved to treat agents as governed identities, with lifecycle controls, ownership records, permission limits and decommissioning rules.

Security researchers warn that merely extending human identity systems to agents is insufficient. AI agents may be short-lived, replicated, delegated to sub-agents and authorised to perform actions across organisational boundaries. That creates problems around transitive delegation, audit trails, temporal validity and permission propagation. A human worker typically has a stable employment relationship and a known device posture. An autonomous agent may have a changing context, a shifting toolset and a task chain that spans multiple systems within seconds.

Web security adds another layer. Cloudflare has expanded tools for identifying and managing automated traffic, including a database for known bots and agents, while also supporting cryptographic verification of automated requests. The approach reflects a wider change in web infrastructure: websites want to distinguish approved AI agents from scraping bots, credential-stuffing tools and abusive automation. For enterprises, that means agents may need to carry verifiable credentials rather than imitate browser behaviour.

The stakes are particularly high in regulated sectors. Agents that read emails, process claims, search personnel files or generate compliance reports may handle personal data at scale. Privacy obligations, data retention rules, automated decision controls and vendor contracts may all be triggered when agents act with limited human supervision. Businesses deploying them must define what data an agent can access, whether it can retain memory, how its output is reviewed and who is accountable when it makes an error.

The article AI agents hit enterprise access roadblocks appeared first on Arabian Post.

The agreement, announced on Monday, allows organisations to build, govern and monitor AI agents on Catalyst while making them available through Microsoft 365 applications including Teams, Outlook and Word. Agents developed within Microsoft Copilot will also be able to flow back into Catalyst, creating a two-way operating model intended to reduce fragmented deployments and improve oversight.

The collaboration gives public-sector bodies and large companies a route to deploy AI agents while keeping sensitive data processed inside the UAE. That point is central to the partnership, as governments, banks, energy groups and regulated companies weigh efficiency gains against concerns over data sovereignty, compliance and auditability.

Inception42, the G42 company focused on sovereign agentic AI, positions Catalyst as a platform for building agents that can operate across workflows, connect to enterprise systems and remain subject to governance controls. Microsoft brings the reach of Copilot and Microsoft 365, already embedded in many corporate and public-sector environments.

Ashish Koshy, chief executive of Inception42, said the integration showed how sovereignty and interoperability could work together. He said Catalyst gave governments and enterprises a platform to build, govern and observe agents, while demand was growing faster than any single platform could address alone.

Amr Kamel, general manager at Microsoft UAE, said customers wanted AI that was powerful, trusted, governed and practical to adopt. He said the ability to process data in-country while embedding AI in everyday tools would help organisations accelerate innovation while retaining control, security and compliance.

The announcement comes as the UAE accelerates efforts to embed agentic AI in public administration. The national agentic AI initiative targets AI support for 50 per cent of federal government operations within two years, while training plans cover 80,000 workers across senior leadership and operational roles.

Abu Dhabi has separately moved to expand Microsoft 365 Copilot across its public sector. Under the Frontier Employee Programme, 26,000 civil servants across 27 entities have received access to Copilot, adding to 9,000 existing users and taking the total to 35,000. The programme is intended to support faster decision-making, improve service delivery and make AI tools part of routine government work.

The latest step builds on a broader G42-Microsoft relationship that has grown steadily since Microsoft agreed a $1.5 billion strategic investment in the Abu Dhabi technology group in 2024. That deal gave Microsoft a minority stake in G42, placed Microsoft president Brad Smith on the G42 board and committed both sides to stronger cooperation on AI infrastructure, cloud services, safety standards and international expansion.

The partnership has also been tied to sovereign cloud development. Abu Dhabi’s agreement with Microsoft and Core42, another G42 company, is intended to create a high-performance sovereign cloud environment capable of handling more than 11 million daily digital interactions across government entities, citizens, residents and businesses. Abu Dhabi’s digital strategy for 2025-2027 includes AED13 billion in investment and a target to automate all government processes.

For G42, the Inception42-Microsoft link strengthens its role in a fast-growing market for sovereign AI, where governments and regulated companies want automation without losing control over data, models and workflows. For Microsoft, the collaboration extends Copilot’s relevance beyond productivity assistance into governed agent ecosystems that can operate across institutional processes.

Agentic AI differs from earlier chatbot-style tools because it can plan, execute tasks, call software tools, retrieve information and take limited actions with reduced human prompting. That capability has increased interest among public agencies and companies seeking productivity gains, but it has also raised questions over accountability, hallucinations, cyber risk, audit trails and the reliability of autonomous outputs.

Industry adoption remains uneven. Many organisations are experimenting with AI agents for coding, customer service, document handling, procurement, finance and internal knowledge management, but production deployments often face barriers around verification, security, data access and integration with legacy systems. Highly regulated sectors are moving more cautiously because errors in automated decisions can carry legal, financial and reputational consequences.

The article G42 and Microsoft widen sovereign AI push appeared first on Arabian Post.

The update rebuilds every edition of the operating system and targets two areas that matter to security researchers, penetration testers, developers and privacy-focused users: better performance on modern hardware and a less cluttered desktop workflow. The release follows Parrot 7.2, which arrived in May, and continues the project’s 2026 shift towards infrastructure polish, reproducible environments and more modular tool management.

A major change is the introduction of optional optimised package repositories for newer processors. Parrot 7.3 adds packages compiled for x86-64-v3 on amd64 systems and ARMv8.2-A on arm64 devices, allowing selected compute-heavy workloads to benefit from newer instruction sets while preserving compatibility for older machines. The project has positioned the feature as opt-in rather than automatic, an important distinction for users who rely on predictable behaviour across mixed hardware fleets.

The performance gains are aimed at shared libraries, language runtimes and heavier utilities used in tasks such as compression, encryption, multimedia processing and numerical computing. Tools such as ffmpeg, Inkscape and NumPy are among the types of software expected to benefit. Compute-bound workloads may see gains of 20 to 50 per cent on supported hardware, though routine desktop actions, shells and network daemons remain tied to conservative baselines to avoid breaking compatibility.

The new package strategy also reflects a cautious approach to security maintenance. Standard security updates retain priority, while optimised packages are delivered through separate repository components to reduce conflicts. Hardware checks are designed to prevent installation on unsupported processors, reducing the risk of users enabling builds that their systems cannot safely run.

The desktop menu has received one of the most visible changes. The older script-heavy menu system has been replaced by two lightweight Go components, parrot-exec and launcher-updater. The move is intended to make application launching faster and tool discovery simpler, particularly for users who do not want a fully loaded system from the start.

Parrot-exec now acts as the unified launcher behind desktop entries. It handles direct execution, privilege escalation and graphical environment variables for X11 sessions, while also supporting on-demand installation. When a tool appears in the menu but is not installed, users can trigger installation from the menu itself, cutting the need to manually search repositories or use the terminal for basic package discovery.

Launcher-updater manages the lifecycle of desktop entries, including generation of templates for tools that are available but not installed, removal of outdated entries and cleanup of duplicate upstream launchers. The result is a more consistent menu structure, with less friction for newcomers and fewer maintenance irritations for experienced operators.

Parrot 7.3 also introduces official Vagrant boxes for Home and Security editions on amd64. The addition is aimed at teams that need identical lab setups for training, testing, continuous integration and repeatable security exercises. The boxes are built from preinstalled qcow2 images, enable SSH access and password authentication, and support quick provisioning through the standard Vagrant workflow.

The browser experience has also been refreshed. Firefox now opens to a lighter Parrot start page built with Vite, offering search options through DuckDuckGo, Qwant and Google, along with links to documentation and project updates. The page is designed to collect no user data, aligning with Parrot’s wider privacy positioning.

The operating system base has been updated with Debian 13.5 “Trixie” components and Linux kernel 7.0.9, improving hardware support and keeping the distribution aligned with current system packages. The update also brings refreshed versions of widely used security and reverse-engineering tools, including Metasploit 6.4.136, Ghidra 12.0.4, sqlmap 1.10.4 and Bettercap 2.41.5.

Parrot’s Home edition remains focused on daily use, privacy and software development, with security tools installable as needed. The Security edition continues to target penetration testing and red-team work, while related editions and delivery formats cover Hack The Box users, Raspberry Pi devices, Docker images, Windows Subsystem for Linux and Debian conversion workflows.

The article Parrot 7.3 sharpens security desktop performance appeared first on Arabian Post.

The feature, launched on 2 July 2026 and switched on by default in Opera’s desktop browsers, targets a fast-growing class of social engineering attacks known as ClickFix. These scams usually begin with a familiar-looking prompt: a failed video player, a broken verification page or a CAPTCHA that asks the user to copy and paste a command to “fix” the issue. The command can then install malware, steal saved credentials or give attackers remote access to the device.

Paste Protect places Opera’s intervention at the point where the attack usually succeeds. Instead of waiting for antivirus software to detect a payload after execution, the browser checks suspicious clipboard activity before the copied content reaches the operating system. When it detects harmful instructions, Opera blocks the copy action and warns the user.

The move reflects a shift in browser security from passive warning systems to direct interruption of risky user actions. For years, browser makers focused heavily on phishing pages, unsafe downloads and malicious extensions. ClickFix has exposed a different weakness: users can be persuaded to run commands themselves, allowing attackers to bypass several conventional defences.

Opera says Paste Protect combines two layers. The first is its existing clipboard hijack protection, introduced in 2021, which guards against attempts to swap copied links, bank account numbers or cryptocurrency wallet addresses with attacker-controlled data. The second is a new injection protection mechanism that looks for malicious scripts and command patterns associated with paste-based attacks.

The feature is aimed mainly at desktop users because ClickFix campaigns often rely on Windows Run, PowerShell, Terminal or command-line interfaces. Attackers typically frame their instructions as routine troubleshooting. A victim may be asked to press a sequence of keys, open a command box and paste what appears to be a harmless verification string. The pasted command may instead download malware, launch a script or contact a command-and-control server.

The urgency behind the release is underlined by the scale of ClickFix activity. Security researchers tracking malware loaders found that ClickFix-style techniques accounted for more than half of loader activity in 2025. The method has been used to distribute information stealers, remote access tools and ransomware-related payloads. It has also spread from criminal groups into espionage-linked campaigns.

Proof-of-concept warnings about pastejacking have circulated for years, but the latest attacks have made the technique more effective by combining trusted visual cues with automated clipboard manipulation. Fake Cloudflare checks, counterfeit CAPTCHA screens and bogus browser error pages are among the most common lures. Many victims do not realise they are executing code because the instruction is presented as a quick fix rather than a security risk.

Opera’s decision to build protection directly into the browser gives it a market distinction at a time when browser security is becoming more competitive. Chrome, Edge, Safari, Firefox and Brave already deploy various protections against unsafe sites, downloads and scripts. Some operating systems and security tools warn users before risky terminal actions. Opera’s claim is that its control sits earlier in the chain, before dangerous content is copied and pasted.

That distinction matters because the browser is often where the deception begins. A compromised website, malicious advertisement or fake support page may not need to deliver a file. It only needs to persuade the user to copy a prepared command. By placing a warning inside the browser interface, Opera is trying to break the trust created by the page before the instruction leaves the web environment.

The feature also reflects a wider trend in endpoint security: stopping actions that appear legitimate but are dangerous in context. Traditional malware detection is less effective when the user appears to be voluntarily launching the command. This has made social engineering more attractive to attackers, especially as defenders have improved detection of suspicious attachments, macros and executable downloads.

There are limits. Paste Protect cannot stop users from manually typing commands, ignoring warnings or using other browsers without comparable safeguards. Developers and advanced users may also need to override blocks when copying legitimate scripts from trusted documentation or repositories. Opera has therefore designed the system to warn and block by default while still allowing more experienced users to control trusted workflows.

The article Opera adds clipboard shield against code traps appeared first on Arabian Post.

The breach, announced on June 30 and updated on July 1, involved systems including Aflac Yorisou Net, a policyholder website used for checking contract details and carrying out account changes. The company said the intrusion was detected on June 25 after abnormal system activity and that related systems were shut down to prevent further access.

The exposed customer data includes names, dates of birth, gender, addresses, telephone numbers, policy numbers, coverage details and premium transfer account information. About 230,000 customers had premium payment account details compromised, including financial institution names, branch names, account types, account numbers and account holder names. National identification numbers and credit card information were not included in the affected data set.

Aflac Japan also said information linked to about 40,000 agencies was exposed, including representative names, agency addresses and telephone numbers. The count includes some former agencies that previously had outsourcing contracts with the insurer. No confirmed misuse of the leaked information has been identified so far, though customers have been urged to report suspicious contact or unusual account activity.

The incident has drawn a formal regulatory response. Japan’s Financial Services Agency issued a report collection order on July 1 under the Insurance Business Act and the personal information protection law, requiring the insurer to report the facts of the incident, customer response measures, root-cause analysis and steps to prevent recurrence. Police have also been notified.

The company’s initial investigation found that the first unauthorised access occurred on June 15 and that multiple attempts followed until June 25. Aflac Japan said it blocked the unauthorised access on the day it was discovered and suspended some systems as a precaution. Claims, benefit requests and customer procedures are still being accepted through call centres and other channels.

Aflac Incorporated, the US-listed parent company based in Columbus, Georgia, said the incident was limited to systems in Japan and that systems linked to its US business were not accessed. The company has engaged external cyber security specialists and said the full scope and financial impact remain under investigation.

The breach is significant because Japan is a core market for Aflac, which built much of its international franchise on cancer and medical insurance products. The company has long held a strong position in Japan’s supplemental health insurance market, supported by agency networks, financial institution partnerships and its policyholder base. Any prolonged disruption to digital policy servicing could test customer confidence at a time when insurers are pushing more account management and claims activity online.

The case also highlights the concentration of sensitive information held by life insurers. Policy records can contain identity data, family details, health-related coverage information, payment instructions and long-term contact histories. Even when credit card numbers and national identification numbers are not exposed, the combination of names, addresses, dates of birth, policy details and bank account data can support targeted phishing, impersonation attempts and social engineering.

Cyber attacks on insurers have become a sharper concern for regulators because the sector sits at the intersection of finance, healthcare and personal identity. Insurers store data that may remain useful to criminals for years, unlike passwords or card numbers that can be changed quickly. Breaches can therefore create risks that extend beyond the immediate theft of files, including fraudulent calls, fake premium requests, bogus claims assistance and attempts to trick customers into disclosing additional information.

Aflac has faced cyber scrutiny before. Its US business disclosed a separate cyber security incident in June 2025 involving unauthorised access through social engineering tactics. That incident involved potentially exposed claims, health and personal information. The company later notified affected individuals after completing a review of impacted files. The Japan breach is being treated as a separate incident, with the company stating that US systems were not accessed.

The latest case places pressure on Aflac Japan to restore affected systems without widening exposure, identify precisely which files were accessed and provide timely notices to customers and agencies. The insurer has said it will send apology and notification letters to affected customers in sequence and will issue further updates if the number of affected people or data categories changes.

The article Aflac breach puts Japan customer data at risk appeared first on Arabian Post.

Between 70% and 95% of digital transformation efforts fail. That’s not a technology problem. Boston Consulting Group research reveals roughly 70% of challenges in AI projects stem from people and process issues, not technical ones.

Yet most organisations continue to treat transformation as a deployment challenge, not a communications one.

My Executive Master’s in Digital Transformation and Innovation Leadership from IE Business School gave me a front-row seat to how transformations are architected. Studying Strategic Communications at LSE showed me why so many of them still fail. The two disciplines together changed how I advise approaching this problem”

Here’s what I see happening on the ground: leadership makes a decision, a transformation team is assembled, a roadmap is built, and technology is procured. Then the announcement goes out — a town hall, a slide deck, an all-hands email. And employees, who were never part of the story, are expected to become part of the solution.

It doesn’t work. 54% of employees feel unprepared to handle changes brought by new technologies, and their lack of readiness is a leading cause of resistance and failure.

The real issue is that organisations confuse informing with communicating.

How does strategic communications help CEOs to drive change?

Some leaders think that communication is just a message sent from an organisation to its audience. In reality, it is rather a negotiation of meaning, during which audiences can define, resist, or co-create that meaning.

In the context of AI transformation, your employees are not passive recipients of change. They are an audience with high power and high interest — the group that, according to the power/interest mapping framework, would be placed at the centre of your engagement strategy. These are the people who control daily workflows, institutional knowledge, and — ultimately — adoption. If they don’t understand the WHY, no amount of training budget or change management process will compensate.

The narrative failure at the heart of most transformations

Organisations that fail at transformation typically commit one of two communications errors:

Telling instead of listening — broadcasting the rationale for transformation without fostering genuine dialogue. The dialogic approach is built on mutual respect, genuine listening, and openness to change. It is demanding precisely because it requires organisations to be vulnerable. To hear what employees are actually afraid of. To acknowledge that the transformation may not yet have all the answers.

Leading with logic and skipping emotion. Behavioural change doesn’t happen through information alone. People shift when they are moved. It’s when they can imagine a different future for themselves within the story you are telling, not just for the company. A narrative that explains what AI will do to the business, without showing what it means for the individual, will consistently fail to land.

So, what works: audience-centric transformation communications

Microsoft’s AI transformation offers a useful contrast. Rather than announcing change, Microsoft used structured change management to first build awareness of the problem being solved, then close the communication gaps between leaders, consulting teams, and end users before asking anyone to adopt. They used the ADKAR model to create awareness, gather feedback on communication gaps, and get all layers of the organisation on the same page — ultimately succeeding in shifting the narrative around their product and increasing adoption of complex cloud-based technology.

ADKAR is a change management model developed by Prosci. It stands for five sequential stages an individual must move through for change to succeed:

• Awareness — of the need for change. Why is this happening at all?

• Desire — to support and participate in the change. Awareness alone doesn’t create buy-in; people need a personal reason to engage.

• Knowledge — of how to change. Training, processes, tools.

• Ability — to implement the required skills and behaviours in practice. The gap between knowing and doing.

• Reinforcement — to sustain the change. Without it, people revert.

The model’s core insight is that organisations typically invest almost entirely in Knowledge (training) and Ability (tools), while skipping Awareness and Desire entirely. That’s precisely the communications failure: you can’t train people into wanting something they don’t understand or trust.

The approach maps directly onto what strategic communications frameworks prescribe: start with context, not solution. Define your most important internal audiences. Understand what they stand to lose and gain. Then build a narrative that transports them — emotionally and rationally — into the future state you are asking them to help build.

A practical checklist for CEOs and Chief Transformation Officers:

• What is the master plot of your transformation story? (Survival? Growth? Relevance? All three read differently to different audiences.)

• How does your narrative balance rational case (efficiency, competitiveness) with emotional resonance (career security, purpose, pride)?

• Are employees participants in shaping the narrative, or recipients of a finished story?

• Have you mapped internal stakeholders by power and interest — and invested your deepest dialogue where it matters most?

• What does success look like beyond adoption metrics? How will you track attitude and behavioural change over time?

What is the business case here?

Change management investments yield 5.3x success rates over technology-only approaches, per McKinsey’s meta-analysis of 1,200 initiatives.  The cost of communications is a multiplier on everything else your transformation budget is trying to achieve.

The organisations that will lead in the AI era are the ones that can take their people with them — honestly, inclusively, and strategically.

That starts with treating internal communication as a discipline, not an afterthought.

The writer is Head of Global Corporate Affairs, Sigma Software Group,  Ambassador of IT Ukraine Association in the UK and UAE, winner of Gulf News & Being She Awards as the best in Technology, category – Digital Transformation

The article Why your AI transformation can fail — and it’s not the technology appeared first on Arabian Post.

The rollout covers three models under the GPT-5.6 family: Sol, the most advanced tier; Terra, a lower-cost model designed for everyday work; and Luna, a faster and cheaper option for high-volume use. Access during the preview is limited to selected organisations through the API and Codex, with ChatGPT users excluded until a broader launch expected in the coming weeks.

The restricted release marks one of the clearest signs yet that advanced AI models are moving into a more sensitive regulatory category, particularly when they show stronger ability to assist with vulnerability research, exploitation analysis and long-horizon software tasks. OpenAI has said the initial group of partners was chosen after consultation with the US government, and that participation details had been shared with authorities before wider access.

The company has framed the step as temporary, arguing that prolonged government-controlled access should not become the standard route for deploying powerful AI systems. Its position reflects a wider tension in Washington and Silicon Valley: security agencies want time to assess dual-use capabilities, while developers and enterprises argue that delaying access can also slow defensive work.

GPT-5.6 Sol is being promoted as OpenAI’s strongest model to date. It introduces a “max” reasoning setting intended to give the model more time to work through difficult problems, and an “ultra” mode that uses sub-agents to divide complex tasks. These features are aimed at software engineering, scientific research and security operations where multi-step planning and tool use are increasingly central.

OpenAI says Sol has set a new benchmark on Terminal-Bench 2.1, a test of command-line workflows involving planning, iteration and tool coordination. It has also shown gains in biology-related workflows through GeneBench v1, which measures long-horizon genomics and quantitative biology tasks. Those claims will draw close scrutiny because frontier models are now being judged not only on accuracy and speed, but also on how well they handle sensitive domains without enabling misuse.

Cybersecurity is the area drawing the most attention. Sol is described as OpenAI’s most capable model yet for cyber work, with improvements in vulnerability research and exploitation-related tasks. On ExploitBench, the model is competitive with Anthropic’s Mythos Preview while using about a third of the output tokens. On ExploitGym, a benchmark developed with academic and industry participation, the GPT-5.6 family shows stronger performance as reasoning effort rises.

The core safety argument from OpenAI is that Sol is better at helping users find and fix software flaws than at reliably carrying out end-to-end attacks. The company says the model does not cross the “Cyber Critical” threshold under its preparedness framework. In tests involving Chromium and Firefox, it found bugs and exploitation primitives but did not autonomously produce a verified full-chain exploit under the tested conditions.

That distinction is important but not conclusive. Security researchers have long warned that the same tools used to audit code, reproduce bugs and generate patches can also reduce the cost of offensive operations. The risk increases when models are linked to browsing, coding environments, scanners, exploit databases or other automated tools. Even if a model cannot independently complete an attack in a laboratory test, it may still improve parts of an adversary’s workflow.

OpenAI has added layered safeguards across the GPT-5.6 family. These include model-level refusal behaviour, real-time checks during generation, account-level monitoring, differentiated access and enforcement measures for misuse. The company has also said some biological and cybersecurity requests may be blocked or delayed while additional checks run, an approach that could affect legitimate work by security teams during the preview.

The system card for the release says safeguard testing has involved expert review, external testing and large-scale automated red-teaming. OpenAI has devoted more than 700,000 A100e GPU hours to automated efforts to find universal jailbreaks and plans to continue red-team testing during deployment. The scale of that testing underlines how frontier AI releases are becoming security operations in their own right, not just product launches.

The pricing structure points to a broader commercial strategy once restrictions ease. GPT-5.6 Sol is priced at $5 per million input tokens and $30 per million output tokens, while Terra is priced at $2.50 and $15, and Luna at $1 and $6. Prompt caching carries a 30-minute minimum cache life, with cache writes billed above the uncached input rate and cache reads receiving a discount.

OpenAI also plans to launch GPT-5.6 Sol on Cerebras in July, offering speeds of up to 750 tokens per second for selected customers as capacity expands. That partnership highlights another emerging trend: model capability is increasingly being paired with specialised inference hardware to make advanced reasoning fast enough for enterprise workflows.

The timing places OpenAI in the middle of a broader debate over how the US should review advanced AI systems before public deployment. Other leading labs have faced similar pressure around models with stronger cyber performance. The emerging policy question is whether governments can build repeatable evaluation systems quickly enough to assess serious risks without freezing access to tools that defenders, researchers and companies may need.

The article OpenAI limits Sol launch amid cyber risks appeared first on Arabian Post.

The decision allows more than 100 vetted organisations to resume use of the company’s most advanced cyber-focused artificial intelligence model. The approved users are understood to include major enterprises, government-linked agencies and institutions responsible for defending networks in sectors such as energy, finance, healthcare, transport and communications.

Access had been suspended on June 12 after federal authorities raised national security concerns over the model’s capabilities and possible misuse by foreign adversaries. The intervention marked one of the strongest signs yet that Washington is prepared to treat frontier AI systems not merely as software products, but as strategic technologies requiring direct government oversight.

Claude Mythos 5 is designed for advanced cybersecurity work, including vulnerability discovery, threat analysis and defensive code review. Its earlier preview version was made available through Anthropic’s Project Glasswing, a programme aimed at helping trusted partners identify and fix software weaknesses before they can be exploited. The initiative was expanded this month to roughly 150 organisations in more than 15 countries, subject to security checks.

The reinstatement is narrower than Anthropic had sought. Access remains limited to approved U. S. organisations, while broader availability of Claude Fable 5, the company’s widely released high-end model, has not fully resumed. The restrictions have left developers, corporate customers and security researchers uncertain about the rules governing future access to frontier models.

The June 12 suspension affected both Claude Mythos 5 and Claude Fable 5. Anthropic disabled access more broadly after being told to restrict use by foreign nationals, a requirement that proved difficult to enforce in real time across cloud services, corporate teams and API customers. The company then entered talks with federal officials to restore access for organisations considered essential to national cyber defence.

The approved carveout reflects a compromise between competing policy goals. Officials are trying to prevent powerful AI tools from being used to accelerate offensive cyber operations, while security teams argue that the same systems are needed to defend complex networks against state-backed hacking groups and criminal ransomware crews.

Cybersecurity specialists have warned that blanket restrictions can weaken defence if they prevent trusted teams from using frontier models to triage alerts, analyse malware, test code and patch vulnerabilities. Large infrastructure operators face persistent pressure from espionage campaigns, supply-chain compromise and attacks on operational technology systems. AI tools are increasingly being tested as force multipliers for small teams handling large volumes of telemetry and code.

The model’s strengths have also made it a focus of regulatory concern. Advanced AI systems can assist with legitimate vulnerability discovery, but the same capabilities may help malicious actors identify exploitable flaws, automate reconnaissance or adapt attack methods faster than defenders can respond. That dual-use nature has pushed cybersecurity to the centre of AI governance debates.

Anthropic has positioned Mythos as a controlled-access model rather than a mass-market product. Its public materials describe Mythos 5 as available only to a small group of vetted partners, with the aim of expanding access over time. The company has said access decisions are linked to safety requirements, partner vetting and the model’s higher-risk capabilities.

The federal action follows months of intensifying scrutiny of advanced AI systems. Policymakers have been moving beyond chip export controls towards direct supervision of model deployment, especially where systems show strong performance in cybersecurity, biotechnology, software engineering and autonomous agent tasks. The Mythos case is being watched by rivals, cloud providers and enterprise customers as a test of how far the government may go in controlling frontier AI releases.

The episode has also exposed tensions inside the AI industry. Companies want predictable rules before investing heavily in specialised models, while government agencies want visibility into capabilities that could alter national security risks. Customers, meanwhile, want assurance that access to critical tools will not be interrupted without clear process or notice.

The article Anthropic reopens Mythos 5 for cyber defenders appeared first on Arabian Post.

The technique, named BioShocking by LayerX Security, was tested against five agentic browsers and one browser-based AI plugin: ChatGPT Atlas, Perplexity Comet, Fellou, Genspark Browser, Sigma Browser and Anthropic’s Claude Chrome plugin. The company said all six systems failed to recognise the final stage of the test as a harmful request when the agent had first been conditioned to accept false rules inside a game.

The demonstration centred on a puzzle designed to make the browser agent abandon ordinary logic. The AI was first encouraged to accept that wrong answers were correct, including a scenario in which 2 + 2 had to be treated as 5. After adapting to the game’s inverted rules, the agent was instructed to visit a path that redirected to what appeared to be an employer’s GitHub repository and copy sensitive login material from a text box.

The test took place in a controlled environment using plaintext credentials, but the security implication is broader. Agentic browsers can interact with pages, tabs, forms, repositories, email accounts and internal business tools that a user is already logged into. Once an AI assistant is granted visibility across that session, a malicious page may be able to influence its next steps without breaching the browser in the traditional sense.

LayerX said vendors were notified. Its disclosure record listed OpenAI’s ChatGPT Atlas as fixed after submission on 30 October 2025, while Perplexity’s Comet was marked as closed or ignored after a 20 October 2025 submission. Fellou, Genspark Browser and Sigma Browser were listed as having given no response after 30 October 2025 submissions, while Anthropic’s Claude Chrome plugin was marked as “patch failed” following a 26 January 2026 submission.

The episode adds to a widening security debate around AI browsers, which are being promoted as a shift from passive browsing to task completion. Instead of simply showing web pages, these tools can summarise sites, compare products, draft messages, fill forms, open links and take multi-step actions. That capability has made them attractive to consumers and enterprises, but it also expands the attack surface because the assistant may act with the same access as the signed-in user.

Prompt injection has become one of the central risks in this category. Unlike conventional malware, it may not require a file download, a malicious executable or an exploit chain against browser code. Instructions can be hidden in web content, comments, documents, URLs or visual elements that the AI interprets while completing a task. The danger arises when the assistant treats attacker-controlled content as an instruction rather than as untrusted data.

Academic work has warned that tool-using AI agents are especially vulnerable when they handle sensitive information across multiple contexts. Studies of agentic systems have found that prompt injection can reduce task reliability and, in some workflows, cause leakage of personal or operational data. Other work on credential exposure in agent skills has identified risks involving secrets, logs and cross-modal interpretation, where code and natural-language instructions combine to create leakage paths.

The BioShocking test highlights a specific weakness: context manipulation. Safety guardrails generally depend on a model recognising that a request is harmful in the real world. If the system is persuaded that the interaction is fictional, game-like or governed by alternative rules, it may treat restricted actions as harmless moves within that scenario. That distinction matters for browsers because the agent’s actions may still occur inside real authenticated services.

Security teams are likely to scrutinise how much authority AI browsers should receive by default. A browser assistant that can read repositories, open email, inspect documents or copy data from enterprise tools may need stronger permission boundaries than a conventional chatbot. Explicit confirmation before accessing sensitive pages, tighter scoping of agent sessions, better detection of role-play attacks and clearer separation between web content and user commands are emerging as key controls.

The article AI browsers face new credential leak warning appeared first on Arabian Post.

The Beijing-based technology group has opened preliminary discussions with banks on a facility that could run for three years, with an option to extend the maturity to as long as five years. The size, pricing and lender group have not been finalised, and the talks may still change before any formal syndication begins.

The proposed borrowing underlines the scale of ByteDance’s capital needs as it expands beyond social media into generative AI, cloud infrastructure, custom chips and data-centre capacity. The company, founded in 2012 by Zhang Yiming and Liang Rubo, has built one of the world’s most profitable private technology platforms through TikTok, Douyin, Toutiao, CapCut and a growing suite of AI products.

A $20 billion transaction would rank among the largest offshore loans by a privately held technology company and would signal strong appetite among banks for exposure to ByteDance despite continuing political and regulatory scrutiny around TikTok. The company remains privately owned, giving lenders limited public financial disclosure compared with listed peers, but its cash generation and global reach have made it one of the most closely watched borrowers in Asia’s technology sector.

ByteDance has been accelerating spending on AI computing power as competition intensifies with Alibaba, Tencent, Baidu, DeepSeek and global rivals. Its Doubao chatbot has become one of China’s most widely used AI applications, while the company has pushed video-generation models, enterprise tools and AI-assisted content creation products into a crowded market. Doubao’s surge during the Lunar New Year period demonstrated ByteDance’s ability to use consumer distribution, entertainment tie-ups and low-friction interfaces to drive mass adoption.

The loan talks come alongside ByteDance’s efforts to secure semiconductor capacity. The company has been working on custom central processing units for AI workloads and has explored external chip-design partnerships. The aim is to reduce dependence on expensive off-the-shelf processors and improve efficiency across data centres serving recommendation systems, chatbots, video generation and advertising tools.

ByteDance’s AI ambitions also depend on access to advanced chips at a time when export controls, supply shortages and rising demand have reshaped the global semiconductor market. The company has pursued overseas computing arrangements, including infrastructure in Southeast Asia, while continuing to invest in model development and inference capacity. This strategy reflects a broader trend among large technology groups, which are moving from purely buying chips to designing parts of their own hardware stack.

The financing push follows a period in which ByteDance’s valuation has remained elevated despite repeated uncertainty over TikTok’s future in the United States. Employee share buybacks over the past year placed the company’s valuation above $300 billion, supported by strong advertising, e-commerce and subscription prospects. TikTok remains a key global asset, but ByteDance’s domestic ecosystem led by Douyin continues to generate substantial revenue and user engagement.

Regulatory risk remains a central consideration for lenders and investors. TikTok has faced national security reviews, data-privacy disputes and ownership pressure in the United States and other markets. The restructuring of TikTok’s US operations reduced one major overhang, but ByteDance continues to operate in a geopolitical environment where technology, data, algorithms and chips are tied closely to state policy.

The company’s offshore borrowing would give it additional flexibility without requiring an initial public offering or a large equity sale. It could use the funds to support AI infrastructure, refinance existing obligations, extend liquidity for investments, or preserve cash for employee share programmes. For a private group of ByteDance’s size, debt financing also allows expansion while avoiding dilution at a time when its valuation is difficult to benchmark against public peers.

Banks are likely to assess the loan against ByteDance’s earnings resilience, its cash balances, the enforceability of offshore structures and the continuing uncertainty around cross-border regulation. Large syndicated loans of this type are often structured with relationship lenders first, before being broadened to other financial institutions if terms and demand are strong.

The article ByteDance plans record offshore borrowing appeared first on Arabian Post.

The project, built with security research firm Trail of Bits and supported by HackerOne, Calif, researchers and maintainers, shifts the focus from simply discovering vulnerabilities to landing tested patches in widely used software. It comes as artificial intelligence tools accelerate bug hunting, creating both opportunities for defenders and a heavier workload for maintainers already dealing with limited resources and large backlogs.

Patch the Planet is designed to pair AI-assisted security research with expert human review. Security engineers use OpenAI’s cyber-capable models and Codex Security to investigate possible vulnerabilities, filter out false positives, develop fixes, improve tests and coordinate disclosure through the channels preferred by each project. Maintainers remain responsible for accepting changes and deciding how fixes are released.

More than 30 open-source projects have committed to take part. Early participants include cURL, NATS Server, pyca/cryptography, Sigstore, aiohttp, the Go project, freenginx, Python and python. org. These projects sit inside critical layers of the digital economy, covering networking, cryptography, language infrastructure, software supply chains and web services used by companies, governments and individual developers.

Trail of Bits has put its security research organisation behind the first phase. Its engineers have worked full-time across 19 open-source projects using Codex and GPT-5.5-Cyber, identifying hundreds of security issues and merging dozens of patches. Some findings remain under coordinated disclosure, meaning technical details will be held back until fixes are available or maintainers complete their remediation process.

The initiative reflects a broader change in cybersecurity. Advanced models can now scan large codebases, reason through possible attack paths, generate proof-of-concept evidence in controlled settings and draft patches. That speed creates a new bottleneck: maintainers must still decide whether a finding is real, how serious it is, whether a patch breaks other functions, and how disclosure should be handled.

OpenAI’s Daybreak update also includes a wider launch of GPT-5.5-Cyber under a limited-access programme for trusted defenders, an updated Codex Security plugin and a partner programme that lets security firms integrate defensive models into their services. Codex Security has scanned more than 30 million commits across more than 30,000 codebases since its research preview began in March. Human reviewers have marked more than 70,000 findings as fixed, while the system has automatically identified more than 500,000 resolved findings.

OpenAI says GPT-5.5-Cyber achieved 85.6 per cent on CyberGym, compared with 81.8 per cent for GPT-5.5, and showed stronger results on other cyber benchmarks. The company is keeping access restricted because the same capabilities that help defenders find and patch flaws could help malicious actors discover attack paths at scale.

Patch the Planet tries to address one of the main complaints from open-source maintainers: automated reports can flood small teams with low-quality findings. The programme requires security researchers to reproduce evidence, remove duplicates, reassess severity and submit only confirmed issues. The model is intended to reduce “slop” vulnerability reports rather than add to them.

The early work has also produced infrastructure intended to outlast the first wave of fixes. Engineers have built fuzzing harnesses, differential-testing systems, historical-CVE analysis pipelines, threat models, expanded test suites and workflows for deduplication and false-positive filtering. In one case, a fuzzing lab covering dozens of entry points, variant builds and platforms was assembled in less than a day, work that would normally take weeks.

Daybreak’s broader testing has also examined operating systems, network software and browsers. The work has included analysis of Linux kernel components, local privilege escalation issues in FreeBSD, a long-standing OpenBSD kernel flaw, vulnerable patterns in dnsmasq, HTTP/2 denial-of-service behaviour affecting major server implementations, and exploitable bugs in browser engines.

The article OpenAI widens Daybreak for software defence appeared first on Arabian Post.

The Santa Clara, California-based company said the Series A-1 extension brings its total funding to $500 million, less than a year after it emerged with a seed round of more than $100 million. The latest round was led by Premji Invest and drew new backing from Nvidia, Salesforce Ventures, Seligman Ventures and Temasek. Existing investors including Maverick Silicon, Mayfield, Prosperity7 Ventures, StepStone Group and Tiger Global also participated.

The fundraising places Upscale AI among a small but fast-growing group of infrastructure start-ups aiming to reshape the data-centre stack beyond graphics processing units. While AI investment has been dominated by demand for accelerators, the company is targeting the fabric that links chips, memory and storage across vast clusters. That layer has become critical as model training and inference workloads stretch across tens of thousands of processors and require predictable, low-latency data movement.

Upscale AI’s pitch rests on open-standard networking designed for AI-native workloads rather than conventional enterprise traffic. The company is building silicon, systems and software that can connect accelerators, memory pools and storage into a high-performance network fabric. Its approach seeks to reduce congestion, packet loss and coordination delays that can leave costly AI chips idle despite heavy capital spending on compute.

The company is led by chief executive Barun Kar and executive chairman Rajiv Khemani. Khemani previously founded Innovium, a data-centre networking chip company acquired by Marvell Technology in a deal valued at about $1.1 billion. That track record has helped Upscale AI attract early confidence from investors familiar with the economics of high-end switching silicon and hyperscale infrastructure procurement.

The latest financing follows a $200 million Series A announced in January, led by Tiger Global, Premji Invest and Xora Innovation, with participation from investors including Intel Capital and Qualcomm Ventures. That round took the company’s total funding above $300 million and established it as a unicorn within months of launch. The new extension doubles its valuation from that level and gives it a larger balance sheet to move products from evaluation to deployment.

Demand for AI networking is accelerating as hyperscalers, cloud providers and specialist “neocloud” operators build larger clusters for training and serving models. The pressure is not limited to compute supply. Data-centre developers are also competing for power, optical components, high-bandwidth memory, switches and interconnect systems. North American data-centre power demand is projected to more than double from 31 gigawatts in 2025 to 66 gigawatts in 2027, underlining the scale of the build-out.

Networking has become a defining constraint because AI workloads behave differently from traditional cloud applications. Large training runs require synchronised communication across many accelerators, with performance hurt sharply when the network becomes uneven or congested. Inference at scale adds another layer of complexity as companies attempt to serve users quickly while controlling energy, memory and routing costs.

That shift is creating openings for suppliers offering Ethernet-based or open networking alternatives to proprietary systems. Spending on data-centre switches used in AI back-end networks is forecast to exceed $100 billion by 2030. Ethernet is gaining ground across both scale-out and scale-up designs because customers want interoperability across accelerators, software stacks and server vendors, although proprietary fabrics still hold an important position in parts of the market.

Upscale AI faces powerful incumbents. Nvidia is expanding its Spectrum-X Ethernet platform alongside its dominant accelerator business. Broadcom remains deeply embedded in switching silicon. Cisco and Arista Networks are strengthening data-centre networking portfolios for AI clusters, while Marvell is active across custom silicon, optical connectivity and infrastructure chips. Start-ups in adjacent areas are also trying to capture value as customers rethink how compute, networking and storage are assembled.

The investor interest reflects a broader recalibration of AI infrastructure spending. The largest technology companies are committing hundreds of billions of dollars to data centres, power contracts and specialised hardware, but investors are increasingly focused on whether that spending can translate into useful capacity and revenue. A cluster filled with accelerators can underperform if the network cannot move data fast enough, making infrastructure efficiency as important as raw chip supply.

The article Upscale AI draws funds for network push appeared first on Arabian Post.

The European Commission’s investigation is focused on whether Meta has done enough under the Digital Services Act to assess and reduce risks to minors, including so-called rabbit-hole effects, addictive design features and weak age controls. The case has become one of the most closely watched tests of the EU’s ability to regulate the architecture of social media platforms rather than only the illegal content that appears on them.

Regulators opened formal proceedings against Meta in May 2024, citing concerns that Facebook and Instagram may stimulate behavioural addiction among children and expose them to harmful or age-inappropriate material through algorithmic recommendations. The probe covers Meta’s risk assessments, mitigation measures, recommender systems, default privacy settings and age-assurance tools.

The next phase is expected to examine more closely whether Meta’s product design creates incentives for prolonged use by minors. Features such as infinite scrolling, personalised feeds, autoplay, notifications and algorithmic amplification have moved to the centre of the European debate, as regulators weigh whether platform engagement tools are compatible with the DSA’s requirement to protect children’s mental and physical wellbeing.

Meta has maintained that Facebook and Instagram are intended for users aged 13 and above and says it has developed tools and policies to protect younger users. The company has pointed to teen accounts, parental supervision features, content restrictions and systems designed to detect and remove underage users. It has also argued that age verification is an industry-wide challenge that cannot be solved by one company alone.

The Commission’s April 2026 preliminary findings raised the stakes. Regulators said Meta had failed to prevent children under 13 from accessing Instagram and Facebook and had not adequately identified, assessed or mitigated the risks linked to such access. Officials said children could bypass age restrictions by entering a false birth date, with no effective controls to verify the information. The EU also criticised tools for reporting underage users as difficult to use and insufficiently effective.

The proceedings could expose Meta to penalties of up to 6 per cent of global annual turnover if breaches are confirmed. For a company of Meta’s scale, that creates a potential multibillion-dollar risk, although EU enforcement usually gives companies a chance to respond, adjust compliance measures and challenge findings before any final decision.

The case forms part of a broader European push to restrict children’s exposure to harmful online design. Commission President Ursula von der Leyen has said the EU is considering tougher rules against addictive and manipulative digital practices through a planned Digital Fairness Act, expected to complement the DSA. She has also raised the possibility of a social media delay or age-related access rules after expert advice is delivered.

The EU has already taken action against TikTok over addictive design, making clear that enforcement will not be limited to Meta. Regulators have identified design patterns such as endless scrolling, autoplay and push alerts as possible drivers of compulsive use, particularly among minors and vulnerable users. The Meta case is likely to determine how far the Commission can extend that logic to the world’s largest social networking group.

Child-safety campaigners argue that voluntary tools have not kept pace with platform incentives. Their concern is that services built around engagement may reward content that keeps children scrolling, even when the material affects sleep, body image, anxiety or exposure to self-harm themes. They also say parental controls often depend on parents discovering problems after harm has already occurred.

Technology companies counter that heavy-handed rules could undermine privacy, limit access to lawful expression and create pressure for intrusive age checks. They warn that stronger age assurance may require sensitive identity data unless regulators and platforms agree on privacy-preserving systems. The Commission has promoted an age-verification blueprint designed to let users prove eligibility without disclosing unnecessary personal information, but implementation remains uneven across member states.

The dispute also carries commercial implications. Meta’s advertising model depends on user attention, personalisation and large-scale engagement. Any order requiring changes to recommender systems, interface design or notification practices could affect product growth and advertising performance in Europe. The company is already operating under strict EU rules on targeted advertising to minors, researcher access, transparency reporting and user choice over personalised recommendations.

National governments are moving in the same direction. Several European countries have debated or advanced limits on social media access for younger teenagers, while Australia’s under-16 ban has become a reference point for policymakers. Britain, France, Norway and other jurisdictions are examining their own approaches, adding to global pressure on platforms to prove that child protection is built into product design rather than added through optional safety settings.

The article EU sharpens Meta child safety case appeared first on Arabian Post.

Cornell Tech researchers Tingwei Zhang, Harold Triedman and Vitaly Shmatikov found that a crafted snippet of about 13 words, planted in user-generated content, could manipulate deep-research systems that gather web material, synthesise it and present confident, citation-backed answers. The technique, named WARP, or Web Agent Retrieval Poisoning, does not require hacking an AI company, breaking into a platform or knowing a user’s exact question. It relies instead on the way research agents repeatedly retrieve the same community pages when answering clusters of related queries.

The study tested three open-source research systems, STORM, Co-STORM and OmniThink, using an ethical simulation framework that avoided altering live web pages. A single poisoned URL produced conditional mention rates of 38 to 51 per cent when the affected page was retrieved, while targeting several URLs lifted the rate to 42 to 62 per cent. In a full-content setting, where the planted material formed less than 4 per cent of the retrieved page, mention rates still reached 30 to 53 per cent.

The risk is acute because consumer questions often lead AI tools to community forums. Queries about restaurant choices, dating apps, cryptocurrency investments or subscription cancellation services tend to draw from Reddit, Wikipedia, Quora, YouTube and similar sites, where ordinary users post informal advice. The Cornell paper found that 17 to 23 per cent of all retrieved URLs in the tested systems came from user-generated platforms, and that an individual community page could appear in up to 48 per cent of queries within the same topic cluster.

That repetition creates a concentrated target. A scammer seeking to promote a bogus service does not need to dominate the web. A strategically placed comment on one thread that research agents already retrieve can be enough to place a fictional name into an AI-generated report. The fabricated examples in the study included a fake Austin restaurant called Sol Azteca, a made-up dating app called SilverPath and a fictitious cryptocurrency presented alongside established digital assets.

The findings also sharpen concerns about commercial AI research products. The researchers did not conduct end-to-end poisoning experiments on ChatGPT Deep Research or Gemini Deep Research because doing so would have required manipulating the live web or observing server-side retrieval that is not externally visible. Instead, they examined how often these tools cite user-generated content during normal use. OpenAI Deep Research cited such material in 3 of 748 reviewed citations, a rate of 0.4 per cent. Gemini Deep Research cited it at 12.1 per cent across the tested topics, suggesting greater exposure to the same structural weakness.

The vulnerability sits at the intersection of retrieval-augmented generation and generative engine optimisation. Retrieval-augmented systems are designed to improve accuracy by consulting current web sources rather than relying only on training data. But when those sources include writable public forums, the system’s strength becomes an opening for manipulation. Generative engine optimisation, a fast-growing marketing practice aimed at influencing AI answers, gives commercial actors an incentive to seed the web with phrases that models are likely to retrieve and repeat.

The Cornell work suggests that conventional defences remain inadequate. Blocking user-generated platforms can stop this class of attack, but it also strips AI research tools of detailed first-hand material that often makes their answers useful. Screening retrieved text before it enters the system was less effective because the poisoned snippets were crafted to read fluently and naturally. Output filtering also struggled because a fake recommendation can appear plausible when it is placed among genuine products or services.

The issue differs from older search-engine spam in one important respect. Search engines usually present users with ranked links, leaving room for scepticism and comparison. Deep-research agents compress multiple sources into a single narrative, often with a tone of authority that can make weak or planted evidence appear more settled than it is. For users, the danger is not merely seeing a bad link but receiving a polished recommendation for something that does not exist.

The article Tiny Reddit posts expose AI search risk appeared first on Arabian Post.

The announcement, made through Trump’s social media account, has not yet been publicly confirmed by either Apple or Intel. Both companies have declined to give detailed comment, leaving open questions about the scale, timing and type of chips that may be covered by the arrangement. Even so, the statement immediately lifted investor expectations that Intel could secure one of the world’s most demanding chip customers after years of trying to build a credible contract-manufacturing business.

Intel shares rose sharply after the statement, with the market treating Apple’s possible participation as a major validation of the company’s manufacturing roadmap. Apple’s stock moved only slightly, reflecting investor caution over whether any initial work with Intel would involve high-volume processors or a narrower set of less critical components.

The potential agreement comes after more than a year of discussions between Apple and Intel over chip production for Apple-designed devices. The talks have been encouraged by the Trump administration as part of a wider industrial policy drive to expand advanced chipmaking capacity in the United States and reduce dependence on Asian manufacturing hubs.

Apple designs the processors used in iPhones, iPads and Macs, but relies heavily on Taiwan Semiconductor Manufacturing Company for production of its most advanced chips. The company moved away from Intel processors in Macs after launching its own M-series chips in 2020, a transition that improved performance and energy efficiency across its computer line-up. A new Intel arrangement would therefore not mark a return to Intel chip designs, but a possible foundry role in producing Apple-designed silicon.

For Intel, the stakes are higher. The company has been trying to reposition itself as a manufacturing partner for outside customers while also competing in the processor market. Its foundry division has required heavy investment and has struggled to prove that it can match the consistency, yields and delivery discipline of TSMC, which remains the dominant producer of cutting-edge chips for Apple, Nvidia, AMD and other major technology groups.

The US government took a 10 per cent stake in Intel last year after converting nearly $9bn in federal support into common stock, making Washington an unusually direct financial participant in the company’s turnaround effort. The administration has since used Intel’s domestic manufacturing base as a central plank in its technology strategy, pressing large chip buyers to consider US production capacity where possible.

Intel has also been promoting its 18A and 18A-P manufacturing processes as proof that its technology is becoming competitive again at the leading edge. The 18A-P process has entered risk production, an early stage used to validate performance and manufacturing readiness before broader customer adoption. The node is designed to deliver better power efficiency and performance, features that are crucial for mobile devices, laptops and artificial intelligence workloads.

Apple’s likely approach would be measured. The company is known for tight control over product quality and supply chain risk, and it is unlikely to move its most commercially sensitive chips away from TSMC without extensive testing. Analysts expect any first Intel-manufactured Apple chips to begin with limited volumes or selected product lines before moving into broader production, depending on yield, cost and delivery performance.

The possible deal also fits Apple’s wider effort to diversify supply chains. The company has expanded US semiconductor commitments through TSMC’s Arizona facilities and other domestic suppliers, while continuing to rely on a global network for assembly, components and logistics. Trump has separately criticised Apple’s overseas manufacturing footprint and pushed the company to make more products for the US market within the country.

The political significance of the Intel claim is considerable. Advanced chips have become central to national security, artificial intelligence, cloud computing and consumer electronics. Washington’s challenge has been to convert subsidies, tax incentives and political pressure into commercially viable production that can compete with Taiwan, South Korea and other established semiconductor centres.

A confirmed Apple order would give Intel a flagship customer and could help attract other companies weighing whether to trust its foundry operation. It would also offer Apple another option at a time when demand for high-end semiconductor capacity is being strained by artificial intelligence chips and next-generation consumer devices.

The remaining uncertainty lies in execution. Apple has not disclosed which chips Intel may produce, when production could begin or whether the arrangement is binding. Intel has not provided customer-specific details tied to its newest process nodes. Until those terms are clarified, the announcement remains a politically significant signal rather than a fully defined supply chain shift.

The article Apple Intel chip plan boosts US foundry push appeared first on Arabian Post.

The platform, announced on June 17, 2026, is available in gated preview and begins with code vulnerabilities, covering first-party and third-party code before AWS expands it to other areas of security. It uses multiple frontier AI models, assigning different models to tasks where they perform best, rather than relying on a single system.

Continuum marks a sharper move by AWS into agentic cybersecurity, where AI systems do more than detect problems. The platform is intended to reason across a customer’s environment, determine which vulnerabilities pose genuine business risk, test exploitability in isolated conditions and then recommend or apply fixes within customer-defined limits.

The launch comes as vulnerability management has become one of the most pressured areas of enterprise security. Software teams are shipping faster, AI-assisted development is increasing code volume, and security teams face a growing stream of findings from scanners, cloud tools, open-source packages and external advisories. Public vulnerability databases are also under strain, with nearly 42,000 CVEs enriched in 2025, 45 per cent more than any previous year, while submissions continued to outpace processing capacity.

AWS is positioning Continuum as a response to that shift. The company says the old model of collecting telemetry, storing it and reviewing dashboards is no longer sufficient when AI models can identify flaws and map complex attack paths at machine speed. The harder problem for customers is deciding which alerts matter, confirming exploitability and fixing the flaw without lengthy coordination between security, engineering and operations teams.

Continuum works in four continuous phases. The discovery phase ingests existing vulnerability backlogs and conducts its own scans across the customer environment. The prioritisation phase evaluates whether an affected component is deployed, reachable, part of a production path and significant to the business if compromised. The validation phase attempts to separate real exposures from false positives by producing reproducible proof in a sandbox. The remediation phase assesses compensating controls and recommends a network change, policy adjustment or code patch.

AWS says the system can also provide blast-radius visibility and rollback paths where feasible, a critical feature for large enterprises wary of automated fixes that may disrupt production systems. Continuum starts in what AWS calls learn mode, keeping a human in the loop and showing the reasoning behind each recommendation. Customers can then move selected categories into enforce mode, allowing more automated remediation under guardrails they define.

The platform incorporates capabilities previously associated with AWS Security Agent. Penetration testing and code scanning are now part of Continuum as Continuum penetration testing and Continuum code scanning, with code scanning still in preview. AWS is also previewing Continuum threat modelling, which can generate STRIDE-based threat models from design documents or source code.

Continuum’s model-agnostic design reflects an emerging pattern in enterprise AI platforms. Instead of building around one foundation model, providers are increasingly using orchestration layers that choose between different frontier models for specialised tasks. For security teams, that could mean using one model to inspect code, another to reason through exploit paths and another to draft remediation steps.

The approach also reflects the growing overlap between offensive and defensive AI. Security researchers have shown that frontier models can help inspect code, reproduce vulnerabilities and generate exploit evidence, but they can also produce false positives or miss vulnerabilities in realistic attack settings. That makes AWS’s emphasis on sandbox validation and staged trust central to whether customers see Continuum as a productivity tool or a source of new operational risk.

The stakes are rising as vulnerabilities move from a compliance concern to a core business risk. Exploited software flaws have played a growing role in breach investigations, while attackers are using automation to reduce the time between disclosure and exploitation. For large organisations, the volume of alerts often exceeds the capacity of human analysts to test and patch every issue manually.

AWS is initially working with select design partners including Capital One, MongoDB, Rivian and Robinhood, indicating that the first wave of adoption is likely to come from technology-intensive companies with large codebases, mature cloud operations and high regulatory exposure. Financial services, automotive and software companies are natural early targets because they combine complex application estates with strict security obligations.

The launch also intensifies competition among cloud and developer platforms seeking to embed AI into software security workflows. Microsoft, Google, GitHub and specialist security vendors are all pushing tools that promise faster code review, threat detection and remediation. AWS’s advantage lies in its access to cloud infrastructure context, permissions, network topology and customer security data, though that same depth will put scrutiny on data handling, model governance and customer control.

The article AWS pushes AI deeper into code security appeared first on Arabian Post.

The European Union’s tender framework for AI gigafactories now allows bidders to propose facilities that can grow in stages rather than committing immediately to the largest scale originally envisaged. The shift reflects mounting concerns over electricity supply, grid connections, financing gaps and the ability of governments and private investors to deliver massive AI computing hubs on a compressed timetable.

The bloc’s original ambition centred on up to five AI gigafactories backed by a €20 billion public-private financing facility. These centres were designed to house more than 100,000 advanced AI processors each, with the computing power needed to train frontier models and support strategic industrial applications. The updated approach keeps that broad objective intact but gives consortia more room to start with lower power capacity and expand as demand, funding and permits mature.

The change marks a pragmatic recalibration rather than a formal retreat from Europe’s AI sovereignty agenda. Officials still want to reduce dependence on US cloud providers, strengthen domestic compute capacity and support start-ups, universities and industry groups that lack access to large-scale AI infrastructure. Yet the new tender shape acknowledges that Europe’s energy and planning systems are not yet aligned with the size and speed of the build-out seen in the US, where hyperscalers have committed tens of billions of dollars to AI campuses.

The Commission’s AI Continent plan sets a target of mobilising €200 billion for AI investment and tripling the bloc’s data centre capacity within five to seven years. The gigafactory programme sits alongside 19 AI factories already selected or operating across Europe, which are built around supercomputing capacity under the EuroHPC framework and intended to provide access for smaller companies, researchers and public-sector users.

The smaller-scale tender option also responds to a sharp distinction between political ambition and market execution. Last year, 76 expressions of interest were submitted across 16 member states and 60 potential sites, suggesting strong early appetite. Since then, the field has been shaped by harder questions: who pays for the chips, how much power can be guaranteed, whether national governments can commit co-financing, and how quickly the sites can secure environmental, grid and land approvals.

Power availability has become one of the most difficult constraints. AI data centres require dense, reliable electricity supplies and advanced cooling systems, while several European markets are already struggling with grid congestion and high industrial energy prices. Operators also face scrutiny over water use, renewable energy procurement and whether public funding should support infrastructure that may benefit a small group of large technology users.

The new model lets bidders choose between capital-support and off-take structures, with public authorities receiving a share of computing capacity over a five-year period. It also requires proposals to spell out the phasing of investment, public support requested, maximum capacity and financial ceilings. That structure is designed to limit the risk of overpromising while giving governments clearer control over what public money buys.

Several countries are still moving aggressively. Spain has approved €719 million for an AI gigafactory project and hopes to align it with EU financing. France has been positioning itself as a major AI infrastructure base, helped by nuclear power, state support and private-sector plans that include multibillion-euro data centre and research campuses. Germany’s telecoms and industrial groups are also exploring large AI data centre projects tied to European funding.

The competitive backdrop remains unforgiving. Europe has strong research institutions, open-source communities and industrial users in sectors such as healthcare, manufacturing, automotive, finance and climate modelling. Its weakness lies in access to large and affordable compute. Three non-European hyperscalers control more than 70 per cent of the region’s cloud market, while the share held by European providers has fallen from 29 per cent in 2017 to around 15 per cent.

The article Europe trims AI data centre blueprint appeared first on Arabian Post.

The agreement, announced on 4 June, is the largest infrastructure commitment in Pinterest’s history and places AWS custom silicon at the centre of the company’s next phase of product development. Pinterest plans to increase its use of AWS Trainium for training and running large language models and vision-language models, while widening deployment of Graviton processors, which already support about a third of its compute infrastructure.

The deal gives Pinterest added capacity for AI workloads at a time when image-led search and automated advertising are becoming central to the company’s growth strategy. Its systems must interpret billions of images, connect them with user intent and turn browsing activity into actionable recommendations for fashion, home design, food, beauty and retail categories.

Matt Madrigal, Pinterest’s chief technology officer, said the company was “heavily investing in AI to make discovery more personal, visual and actionable” for users. He said the expanded AWS commitment would provide “compute flexibility, hardware optionality and infrastructure efficiency” and help improve consumer experiences and advertiser performance through proprietary models and open-source models.

Pinterest and AWS have worked together since 2010, helping optimise large-scale data lake infrastructure and core service reliability. The agreement extends that relationship across AI model training, inference and platform infrastructure, deepening AWS’s role as Pinterest’s preferred cloud services provider.

The company also plans to continue shifting from traditional EC2-based environments to a Kubernetes-based architecture on Amazon Elastic Kubernetes Service. The migration is intended to improve developer velocity, reliability and efficiency as AI-assisted products become more data-intensive.

Pinterest’s AI push is linked to its broader effort to define itself less as a conventional social network and more as a visual search and shopping engine. Its proprietary Taste Graph helps connect user interests with personalised recommendations, while transformer-based models and multimodal systems have improved the way the platform ranks, retrieves and serves content. Pinterest Assistant, a conversational discovery feature powered by open-source vision-language models, forms part of that shift.

The agreement follows a stronger start to 2026 for Pinterest. The company reported first-quarter revenue of US$1.008 billion, up 18 per cent year on year, with global monthly active users rising 11 per cent to a record 631 million. It posted a GAAP net loss of US$74 million, adjusted EBITDA of US$207 million, operating cash flow of US$328 million and free cash flow of US$312 million for the quarter.

Bill Ready has argued that Pinterest occupies a distinctive position because users often arrive with purchasing or planning intent rather than purely for entertainment. That distinction has become more important as advertisers demand measurable returns and platforms race to connect search, recommendations and commerce. AI-powered products such as Pinterest Performance+ are designed to automate campaign optimisation and make advertising more effective across search and shopping surfaces.

For AWS, the Pinterest commitment adds to a wave of large enterprise cloud deals tied to AI infrastructure and custom chips. Snowflake signed a US$6 billion AWS agreement in May covering AI and data workloads, including Graviton processors. Amazon has also stepped up financing for its infrastructure buildout, securing a US$17.5 billion delayed-draw loan facility in June as hyperscalers pour capital into data centres, chips, networking, energy and cooling systems.

The article Pinterest deepens AWS cloud push for AI discovery appeared first on Arabian Post.

The Santa Clara-based chipmaker priced the investment-grade deal on June 15 after initially targeting at least $20 billion. Orders reached about $85 billion, allowing the company to expand the transaction while keeping borrowing spreads tight. The sale marked Nvidia’s first return to the corporate bond market since 2021, when it raised $5 billion.

The offering was structured across seven tranches, with maturities running from 2028 to 2056. The longest-dated securities put investors behind Nvidia’s AI strategy for three decades, a notable vote of confidence in a company whose processors have become central to training and running advanced machine-learning systems. Goldman Sachs, JPMorgan and Morgan Stanley led the transaction.

The proceeds are earmarked for general corporate purposes, including the repayment and refinancing of outstanding notes. That makes the deal less a distress-driven funding exercise than a move to deepen liquidity, establish a larger credit benchmark and give Nvidia greater balance-sheet flexibility as the AI industry enters a more capital-intensive phase.

Nvidia already has considerable financial strength. Revenue for the quarter ended April 26 reached $81.6 billion, up 85 per cent from a year earlier, while data-centre revenue climbed to $75.2 billion. Gross margin stood at 74.9 per cent, reflecting the company’s pricing power and the rapid shift of computing budgets towards accelerated processing.

The company held $50.3 billion in cash, cash equivalents and marketable debt securities at the end of the quarter, along with $30.2 billion in marketable equity securities. It had $8.5 billion of senior notes outstanding before the new deal and no borrowings under a $25 billion commercial paper programme. The new bonds will substantially increase its debt load but still leave leverage modest relative to earnings and cash generation.

The timing points to a broader change in technology finance. The AI boom is moving from a software-led story into a heavy infrastructure cycle requiring chips, networking equipment, electricity, cooling systems, cloud capacity and long-term supply agreements. Nvidia is not spending on hyperscale data centres in the same way as its largest customers, but it is investing heavily in product roadmaps, engineering capacity, supply commitments and strategic stakes across the AI ecosystem.

That spending backdrop has made corporate debt a more attractive tool for even the richest technology groups. Borrowing allows companies to preserve cash, refinance older obligations and avoid issuing equity after large share-price gains. Nvidia’s shares rose more than 3 per cent on the day of the bond pricing, suggesting equity investors viewed the transaction as a sign of confidence rather than balance-sheet strain.

The company’s annual product cadence is another factor behind the bond market’s appetite. Nvidia has shifted to releasing new AI chip families at a faster pace, with Blackwell systems driving demand and the Vera Rubin platform forming the next major architecture. Customers include cloud providers, AI laboratories, sovereign computing projects, enterprises and industrial users seeking higher-performance training and inference capacity.

The deal also reflects investors’ search for high-quality corporate paper at a time when US Treasury yields remain elevated and credit markets are absorbing large technology-sector issuance. Nvidia’s strong rating profile, dominant market position and expanding cash flow make its bonds attractive to insurers, pension funds and asset managers seeking long-duration exposure to the AI economy.

Risks remain significant. Nvidia’s filings point to export controls, China-related restrictions, supply-chain dependence and large purchase commitments as continuing pressures. The company has said it is effectively blocked from competing in China’s data-centre compute market under current rules, while shifting regulations could affect sales in other regions. Competition from custom chips designed by cloud companies and processors developed outside the United States could also test Nvidia’s margins over time.

The debt sale therefore lands at a pivotal moment: demand for Nvidia’s chips remains exceptionally strong, but the financing demands of AI are becoming too large to sit only on corporate cash flow. By locking in long-term funding out to 2056, Nvidia has given itself more room to manage refinancing, investment and strategic commitments while investors have taken a direct credit position on the durability of the AI buildout.

The article Nvidia debt rush sharpens AI market bet appeared first on Arabian Post.

The article Telegram ban blows open India’s exam-security crisis appeared first on Arabian Post.

The initiative brings together BNY, Cisco, Cloudflare, Corridor, DepthFirst, Docker, JPMorganChase, Kyndryl, LTM and PwC, with Chainguard acting as the coordinating platform. Athena is already operational with more than two dozen participating organisations, having processed over 20,000 vulnerability findings and generated more than 2,000 patches across 500 open-source projects. Its first coordinated disclosure wave is scheduled for July.

The move marks a shift in software security practice from disclosure-led patching to pre-embargo defence. Traditional coordinated disclosure was built around a slower sequence: a flaw is found, maintainers are notified, a patch is prepared, an advisory is published and users are expected to update. Athena is designed for a faster environment in which advanced AI models can read large codebases, trace dependencies and identify chained zero-day vulnerabilities before many maintainers or vendors can respond.

Chainguard chief executive and co-founder Dan Lorenc said the exploit window has changed sharply. “The time to exploit has gone negative — exploits now land before a flaw is ever disclosed,” he said. “Athena’s whole job is to make the time to remediate even more negative, so the fix is already in place before the vulnerability is public. No one company can get ahead of this alone, and orchestrated defence is the only answer.”

Athena will collect vetted findings from member organisations, including those produced through frontier AI security programmes such as Anthropic’s Project Glasswing and OpenAI’s Daybreak. The coalition then deduplicates reports, enriches them with technical context, traces when a weakness entered the code, checks whether the flaw has already been repaired upstream and identifies related patterns across other projects.

The platform’s most sensitive work happens before public disclosure. Affected projects can be rebuilt as private, hardened versions and made available to members through Chainguard Libraries while an embargo remains in place. The coalition also intends to harden entire libraries against classes of flaws rather than treating each AI-discovered bug as an isolated defect. That approach is aimed at preventing a stronger model or a malicious actor from finding the next variant in the same code path.

Network and platform partners have a central role because many organisations cannot patch critical systems within hours. Athena’s model allows infrastructure providers and security vendors to prepare traffic-level rules, detection signatures, virtual patches and platform-side blocks before details become public. The aim is to reduce exploitation risk even where a clean software patch is not yet deployed.

The coalition is also seeking to avoid a fragmented response in which cloud providers, software vendors and security teams privately fork the same libraries with separate patch sets. That fragmentation can leave no shared record of what has been fixed, where fixes have diverged and which systems remain exposed. Athena’s clearinghouse structure is intended to create a common technical pipeline while allowing members to retain control over what they share and on which embargo timeline.

The timing is significant for banks and large enterprises, which depend heavily on open-source components across payment systems, trading platforms, cloud infrastructure, artificial intelligence tools and internal development pipelines. Commercial software audits show that open-source code now makes up a majority of modern codebases, while dependency chains have become deeper and harder to monitor. Security teams are also dealing with a sharp rise in malicious packages across public registries, where attackers target developers, build tools and credentials rather than only finished applications.

BNY chief information security officer Dave Robinson said trust in underlying software had become a direct operational issue. “Our clients count on BNY to protect what matters most, including the software behind our systems. As AI speeds up the discovery of vulnerabilities, Athena may help us identify and address risks earlier,” he said.

The article Athena targets AI-led open source exploits appeared first on Arabian Post.

The European Commission is examining the practical consequences for users across the bloc after Anthropic disabled Fable 5 and Mythos 5, saying it had received a US export-control directive barring access by any foreign national, including non-US staff inside the company. The measure, issued on national security grounds, has raised concern that frontier AI services can be withdrawn from allied markets with little warning when Washington classifies them as sensitive technology.

A Commission spokesman said emergency measures should not be discriminatory against partners, a pointed response that stops short of a formal challenge but signals discomfort in Brussels over the breadth of the order. The case lands at a delicate moment for transatlantic technology policy, with Europe trying to tighten AI governance while accelerating domestic computing capacity to reduce reliance on foreign providers.

Anthropic said the directive arrived on June 12 at 5.21pm Eastern Time and did not give detailed reasons for the national security concern. The company said US officials believed they had become aware of a method to bypass safeguards in Fable 5, but argued that the demonstrated technique identified only a small number of known, minor vulnerabilities. It said similar vulnerabilities could be found using other publicly available models without such a bypass.

The order affected two models launched only days earlier. Fable 5 was presented as a broadly available frontier model with strong safeguards, while Mythos 5 used the same underlying model with some safeguards lifted for vetted cybersecurity and research partners. Anthropic said Fable 5 could operate autonomously for longer than previous Claude models, improve productivity in large software projects and handle complex vision, document and analytical tasks.

The suspension has widened a debate over whether advanced AI models should be treated like chips, cryptography or other dual-use technologies subject to export licensing. US policy has already restricted advanced semiconductors and cloud computing access for sensitive jurisdictions, but the Anthropic order pushes the boundary closer to the model layer itself. For enterprise customers, that shift changes risk calculations around procurement, continuity planning and the legal exposure of AI systems embedded in banking, health care, government services and cybersecurity operations.

European officials are weighing the issue against the bloc’s own AI rulebook. The AI Act, which entered into force in 2024, gives the EU a risk-based framework for governing AI systems and general-purpose models, with the European AI Office playing a central supervisory role. Yet the Anthropic case shows that regulation alone cannot guarantee access when the underlying technology is controlled by a company subject to another government’s security directives.

The timing is awkward for Anthropic’s international expansion. The company has been courting regulated industries and had announced partnerships intended to bring Claude systems into large organisations across multiple countries. A directive applying to foreign nationals as a category, rather than to specific hostile actors or sanctioned entities, creates operational complications for multinational clients and vendors with global engineering teams.

Washington’s case rests on the argument that frontier models with advanced cyber capabilities could be misused by foreign military or intelligence services. That concern is not dismissed in Europe, where governments face persistent cyber espionage, disinformation and infrastructure threats. The tension lies in the scope of the remedy. Blocking all foreign nationals, including allies, may protect one risk channel while undermining confidence in US technology platforms as dependable infrastructure.

The episode is likely to strengthen Europe’s sovereign AI agenda. The Commission has already set out plans to mobilise €200 billion for AI development, including €20 billion for up to five AI gigafactories, alongside a network of AI factories intended to support start-ups, industry and research. Those programmes were designed to close Europe’s compute and deployment gap; the Anthropic cutoff adds a strategic-access argument to the economic case.

The article EU questions US clampdown on Anthropic AI appeared first on Arabian Post.

The San Francisco-based company said the accounts were used to produce social media comments, comic strips, slogans and political cartoons for apparent covert influence operations. The activity was assessed as likely originating from China and was removed after internal investigators found the accounts had used OpenAI’s models to support narratives aimed at inflaming domestic disputes in the United States.

The data centre effort, named “Data Center Bandwagon” by OpenAI, focused on complaints that AI facilities were straining power grids, raising utility costs and harming communities. Operators asked ChatGPT to draft English-language posts and images, often while prompting the system in Simplified Chinese and using virtual private networks. The content was then pushed through likely inauthentic accounts posing as Americans.

OpenAI said the operation failed to gain meaningful traction beyond its own activity. That assessment limits the immediate political impact of the campaign, but the case has drawn attention because it shows how foreign operators are experimenting with generative AI to scale narratives around sensitive infrastructure projects. The company said the campaign did not create opposition to data centres; it attempted to exploit concerns that already existed.

A second cluster, called “Tech and Tariffs”, generated material criticising US tariff policy and efforts to dominate technological competition. Some prompts instructed the model to depict President Donald Trump while avoiding references to China’s leader Xi Jinping. The cluster also produced content in multiple languages and was connected to claims that ChatGPT user data had been compromised, an allegation OpenAI rejected as false.

The activity appears to have run from late 2025 into early 2026. One group involved in the data centre campaign was assessed as likely linked to a private technology firm working for provincial-level government clients in China. The operators also sought help with operational tasks, including polishing work reports and discussing ways to avoid detection by social media platforms.

The China embassy in Washington said it was not familiar with the report and rejected what it described as groundless attacks and smears. Beijing has repeatedly said it supports responsible use of artificial intelligence and opposes disinformation, while Washington and technology companies have grown more vocal about covert influence activity tied to strategic competition over AI.

The disclosure comes as data centre construction has become a sharper political issue in the United States. Generative AI systems require vast computing capacity, and the facilities that train and run them need large volumes of electricity, water and land. Global data centre electricity consumption is projected to roughly double by 2030, reaching about 950 terawatt-hours and accounting for around 3 per cent of global demand. AI-focused facilities are expected to grow faster than the wider sector.

US power demand is also rising after years of comparatively flat growth. Electricity consumption is forecast to hit record levels in 2026 and 2027, with commercial use overtaking residential demand for the first time as data centres, manufacturing and electrification lift load on the grid. Local opposition has grown in parts of Virginia, Georgia, Arizona, Texas and other states where communities are questioning water use, noise, land deals, tax incentives and grid costs.

That context makes the OpenAI findings politically sensitive. Supporters of rapid AI infrastructure expansion argue that foreign-linked actors are trying to slow US technological leadership by amplifying environmental and consumer concerns. Critics of the industry counter that foreign manipulation should not be used to dismiss genuine local grievances over power bills, planning transparency and resource consumption.

OpenAI’s findings give limited support to warnings that outside actors are seeking to influence the data centre debate, but they do not show that public opposition is primarily foreign-driven. The company’s own assessment indicates the accounts had little visible effect, reinforcing a broader pattern in which generative AI helps operators produce more content without necessarily producing more persuasion.

The article OpenAI blocks China-linked data centre campaign appeared first on Arabian Post.

The discussions remain fluid and may still change in size, timing or investor composition. If completed near the terms being discussed, the financing would mark another significant jump for a company founded only in 2023 and already seen as Europe’s most credible challenger to OpenAI, Anthropic and Google DeepMind.

The proposed valuation would represent a steep rise from Mistral’s last confirmed funding round, when it raised €1.7 billion at an €11.7 billion post-money valuation in September 2025. That round was led by ASML, the Dutch semiconductor equipment group, which invested €1.3 billion and took an approximately 11 per cent fully diluted stake, making it Mistral’s largest shareholder.

The new capital would give Mistral greater room to expand its model development, data-centre capacity and enterprise AI services at a time when foundation-model companies are under pressure to spend heavily on graphics processors, power supply, engineering talent and global sales. The cost of remaining competitive has risen sharply as AI laboratories race to train larger models, improve reasoning capabilities and integrate agents into business workflows.

Mistral has sought to differentiate itself through open-weight models, enterprise-focused deployment and a sovereignty message aimed at companies and governments seeking alternatives to US-controlled platforms. Its founders, Arthur Mensch, Guillaume Lample and Timothée Lacroix, previously worked at leading AI research organisations, giving the start-up technical credibility from its earliest days.

The company’s commercial strategy has also shifted from building language models alone to creating a fuller technology stack. It has added tools for training, testing and deploying custom AI systems, while pushing products aimed at coding, workflow automation and long-horizon tasks. Its Vibe assistant, unveiled at its AI Now Summit in Paris, is positioned as an agent for multi-step work across inboxes, calendars, research, coding and recurring business processes.

Mistral’s expansion has been accompanied by acquisitions designed to strengthen its infrastructure and industrial capabilities. The company bought Koyeb, a cloud services start-up based near Paris, in February, bringing its 13 employees into Mistral as part of a broader move to control more of the computing layer. It later acquired Linz-based Emmi AI, an Austrian physics AI specialist whose models are used for simulations involving airflow, heat transfer and material stress.

Those deals point to a strategy aimed at engineering-heavy sectors rather than only consumer chatbots. Mistral has been building relationships across semiconductors, automotive, aerospace, energy, banking and defence, pitching itself as a partner for companies that want AI models adapted to proprietary data and deployed under tighter control.

The ASML partnership remains central to that positioning. The semiconductor equipment group said the collaboration would explore AI models across its product portfolio, research and operations, with the goal of improving lithography systems and accelerating engineering work. For Mistral, the alliance linked it to one of Europe’s most strategically important technology companies and gave it an investor with deep knowledge of the chip supply chain.

Infrastructure is becoming an equally important part of the story. Mistral has outlined plans to build and operate more computing capacity in Europe, including data-centre projects in France and Sweden. It has also defended the need for European AI capability in sensitive fields, including defence, arguing that technological dependence on foreign platforms would leave governments and industries exposed.

The company’s ambitions mirror a broader policy push across Europe to build sovereign AI capacity. Governments and large corporations are trying to reduce reliance on American cloud providers and AI labs, while still competing in a sector where US companies have raised far larger sums. Mistral’s fundraising effort comes against that backdrop of strategic urgency, with investors weighing whether Europe can produce a global AI champion with enough capital, chips and customers to endure.

Competition remains formidable. OpenAI, Anthropic, Google and Meta continue to command far larger pools of capital, deeper computing resources and broader consumer reach. Chinese AI developers are also advancing quickly, while open-source communities keep lowering barriers to model access. For Mistral, the challenge is to prove that a European model provider can scale revenue fast enough to justify rising valuations while sustaining the research spending required at the frontier.

The proposed €20 billion valuation would place greater scrutiny on execution. Investors will look for evidence that enterprise contracts, public-sector adoption and industrial partnerships can translate into recurring revenue at scale. They will also assess whether Mistral’s open and sovereign positioning can remain commercially attractive as customers demand stronger performance, lower inference costs and clearer security guarantees.

The article Mistral seeks fresh billions for AI expansion appeared first on Arabian Post.

The move puts AMD more directly against Arm-based suppliers that have built strong positions in embedded and automotive systems through power-efficient designs, licensing flexibility and broad developer ecosystems. AMD’s argument is that a single x86-based system-on-chip combining CPU, graphics and neural processing can give equipment makers enough performance for real-time workloads without adding separate accelerators.

The Ryzen AI Embedded P100 and X100 families combine Zen 5 CPU cores, RDNA 3.5 graphics and XDNA 2 neural processing units. The P100 range is aimed at lower-power embedded systems, including digital cockpits, industrial automation, human-machine interfaces and robotics controllers. The X100 line is positioned for heavier edge AI and physical AI workloads where machines must interpret sensor data, make decisions and act with low latency.

AMD has said the P100 series offers configurations from four to 12 CPU cores and up to 50 trillion AI operations per second, depending on the model and implementation. The chips are designed for long-life embedded platforms, ruggedised operating conditions and virtualised deployments where control, safety partitioning and resource allocation are essential. That makes the line relevant for sectors where product cycles often run for many years, unlike consumer PCs.

Automotive is one of the clearest targets. Vehicle manufacturers are shifting from distributed electronic control units to software-defined platforms that consolidate computing across cockpit displays, driver assistance, infotainment and in-cabin AI features. AMD is seeking to place Ryzen AI Embedded processors in this transition by offering graphics, conventional compute and AI inference on the same silicon, reducing board complexity and potentially lowering system power and cost.

Robotics is another growth area. Warehouse automation, autonomous mobile robots, inspection systems and humanoid platforms increasingly require onboard inference rather than constant cloud connectivity. Latency, data privacy and reliability concerns are pushing more AI workloads to the edge, where processors must handle cameras, lidar, speech, navigation and control loops within tight thermal envelopes. AMD’s x86 pitch is aimed at developers who want PC-class software compatibility alongside dedicated AI acceleration.

The company’s embedded push also reflects a wider effort to diversify AI revenue beyond data-centre GPUs. Nvidia remains dominant in AI accelerators, while Qualcomm, NXP, Renesas, MediaTek and other Arm-linked players continue to compete in automotive, internet-of-things and low-power edge computing. RISC-V is also gaining attention as manufacturers seek open architectures and alternatives to established instruction sets.

AMD’s strength lies in performance computing, graphics and its existing x86 software base. Many industrial and medical equipment developers already use x86 platforms because of operating system compatibility, development tools and long-standing application support. The embedded Ryzen AI strategy seeks to preserve that advantage while addressing a market that increasingly asks for AI inference, image processing and deterministic control in the same package.

The challenge is power efficiency and ecosystem depth. Arm-based processors dominate many embedded categories because they are widely licensed, customisable and used across mobile, automotive and IoT designs. Arm vendors can offer highly integrated chips with radios, sensor interfaces, microcontrollers and safety-certified features tailored for specific markets. AMD will have to persuade manufacturers that x86 performance and software continuity outweigh the power and cost advantages often associated with Arm designs.

There are signs of growing partner interest. Embedded board makers and module suppliers have begun positioning Ryzen AI Embedded platforms for industrial PCs, computer-on-modules, edge gateways and machine-vision systems. These products are critical because many industrial customers buy complete modules rather than chips directly, relying on hardware partners for thermal design, I/O integration and long-term availability.

Medical systems could provide another opening. Imaging equipment, diagnostic instruments and surgical platforms increasingly use AI for image enhancement, segmentation and workflow automation. These devices require predictable performance, strict validation and long support windows. A processor combining graphics and neural acceleration could appeal to manufacturers trying to modernise equipment without adopting a fragmented hardware stack.

AMD’s timing is shaped by the spread of “physical AI”, a term used for systems that apply artificial intelligence to machines operating in the real world. Unlike cloud chatbots or office copilots, physical AI requires fast local decisions, sensor fusion and reliable control. Cars, robots and factory equipment cannot always wait for remote servers to process data, making edge AI silicon a strategic battleground.

The article AMD pushes x86 deeper into edge AI appeared first on Arabian Post.

The demonstration showed the Windows version of Half-Life running on a Dell OptiPlex desktop fitted with an Intel Core i5-2400 processor and an NVIDIA GeForce 8400 GS graphics card. The significance lies less in the age of the 1998 first-person shooter than in what it proves about ReactOS: the system is now able to handle a demanding mix of legacy Windows application calls, graphics driver behaviour and OpenGL rendering well enough to move beyond launch screens and into live gameplay.

Half-Life remains a useful compatibility test because it sits at the intersection of older Windows gaming, the GoldSrc engine and hardware-accelerated OpenGL. Earlier ReactOS experiments had shown the game starting or reaching partial states, while Half-Life 2 had appeared in limited tests through workarounds. This run points to a deeper level of stability in the graphics stack and driver path, particularly after months of work on GPU driver compatibility.

The milestone follows a March 2026 advance in ReactOS graphics support involving work around KMDF, the Kernel-Mode Driver Framework, and WDDM, the Windows Display Driver Model. That work helped improve compatibility with a wide range of Windows XP and Windows Server 2003-era graphics drivers, including hardware from Intel, NVIDIA and AMD. Tests during that phase covered cards and chipsets such as Intel GMA 945, NVIDIA GeForce 8800 GTS, GTX 750 Ti, Quadro 1000M, GTX Titan X and AMD Radeon HD 7530G.

For ReactOS, graphics support has long been one of the hardest barriers to wider usability. The project aims to run Windows applications and drivers natively, rather than through a compatibility layer such as Wine, but that ambition requires a close reimplementation of complex kernel, driver and user-interface components. Modern display drivers depend on interactions between user-mode libraries, kernel-mode components and graphics hardware, while older applications often expect behaviour inherited from Windows 2000, XP and Server 2003.

WDDM is especially important because it reshaped Windows graphics from Vista onward, moving more responsibility for GPU management away from older Win32k paths and into the DirectX graphics kernel subsystem. ReactOS has historically targeted an NT 5. x-style environment, broadly aligned with Windows Server 2003, but its developers have had to engage with later driver architecture to keep the system relevant for real hardware.

The Half-Life breakthrough also underlines why games are often used as practical stress tests for operating systems. A game may expose timing issues, driver faults, input bugs, file-system assumptions and graphical glitches faster than productivity software. GoldSrc’s OpenGL pipeline requires the system to load and coordinate graphics components correctly, while maintaining the application compatibility needed by a commercial Windows title from the late 1990s.

ReactOS remains an alpha-stage operating system and is not positioned as a daily replacement for Windows. Installation, hardware support and application reliability can vary sharply by machine. The successful Half-Life run therefore does not mean modern Windows gaming is suddenly viable on ReactOS, nor does it remove the larger engineering burden facing the project. It does, however, show measurable progress in an area where failures have historically been visible and difficult to mask.

The project’s broader appeal lies in its unusual position within open-source computing. Linux, BSD and other free operating systems have mature ecosystems of their own, while Wine and Proton have become central to running Windows games and applications on Linux. ReactOS is different: it seeks binary compatibility with Windows applications and drivers at the operating-system level, using a clean-room approach and open-source development.

The article ReactOS passes Half-Life graphics hurdle appeared first on Arabian Post.

The programme would move part of the AI computing race beyond land-based data centres, where power shortages, grid constraints, cooling costs and local opposition have become major obstacles for technology companies. The first demonstration missions are expected to test whether high-performance chips can operate reliably in space, communicate through optical links and use solar power at a scale that could support commercial AI workloads.

The plan places SpaceX at the centre of a fast-expanding contest between launch providers, cloud companies, chipmakers and AI developers. Musk has argued that orbital computing could benefit from abundant solar energy and avoid many of the physical limits facing terrestrial data centres. SpaceX’s existing Starlink network gives the company operational experience in building, launching and managing large satellite constellations, but AI data centres in orbit would demand far greater power, thermal control and hardware resilience.

The first orbital AI satellite is expected to rely on Nvidia chips and deliver computing capacity comparable to a high-end AI server rack on Earth. The spacecraft design under discussion would use solar arrays for power, radiators for heat management and laser links or Starlink connectivity for data transfer. Musk has said the system can draw heavily on technology already developed for Starlink, though the economic case remains unproven.

SpaceX has sought US regulatory clearance for a future constellation that could ultimately include up to one million satellites operating as orbital data centres. Such a scale would far exceed current satellite deployment levels and would intensify scrutiny from regulators, astronomers, environmental researchers and rivals concerned about orbital congestion. The company’s proposal frames the system as a way to meet surging AI demand without relying solely on Earth-based power grids.

The timetable is tied closely to Starship, SpaceX’s fully reusable heavy-lift rocket. Starship is central to Musk’s cost assumptions because large-scale orbital computing would require frequent launches of heavy satellites, replacement hardware and supporting systems. Delays in Starship development would therefore slow any commercial roll-out, even if early demonstrations succeed.

The project is also becoming part of SpaceX’s broader investor pitch as the company seeks to expand beyond rockets, Starlink broadband and government launch contracts. AI infrastructure has become one of the most valuable themes in global markets, with cloud providers spending heavily on chips, power agreements and data-centre campuses. SpaceX is positioning itself as a company that can combine launch dominance, satellite manufacturing, power from space and high-speed communications into a new computing platform.

Google is pursuing a parallel effort through Project Suncatcher, which aims to test solar-powered satellites equipped with its Tensor Processing Units. That mission, planned with Planet Labs around 2027, is designed to examine how machine-learning hardware performs in orbit and whether satellites can work together through optical inter-satellite links. The overlap suggests that space-based compute is moving from speculative theory into early engineering trials, though commercial viability remains uncertain.

The technical hurdles are significant. AI chips generate intense heat, and space offers no air or water cooling systems of the kind used in terrestrial data centres. Heat must be radiated away, requiring large surface areas and careful design. Radiation can damage electronics, while hardware failures in orbit are far harder to repair than failures in a warehouse-scale data centre. Latency, data transfer costs and cybersecurity will also shape whether customers see orbital AI compute as practical.

The regulatory questions may prove just as difficult. A million-satellite architecture would raise concerns about collision risks, space debris, radio-frequency coordination and the brightness of objects crossing the night sky. Astronomers have already warned that very large constellations can interfere with observations, while space-safety experts have urged stronger traffic-management rules as low Earth orbit becomes more crowded.

The article SpaceX targets orbital AI test launch appeared first on Arabian Post.

The change means millions of users who bought relatively modern Apple Watches will remain on watchOS 26 unless they move to newer hardware. The supported list for watchOS 27 starts with Apple Watch Series 9 and extends to Series 10, Series 11, Ultra 2, Ultra 3 and SE 3. The cut-off is more severe than many previous watchOS transitions because it removes models that were still being sold or widely promoted within the past few years.

Apple previewed watchOS 27 at its Worldwide Developers Conference on 8 June, positioning the update as part of a wider software reset built around Siri AI, Apple Intelligence, health features and tighter links between the Watch and iPhone. The developer beta is available now, with a public beta expected in July and the full release planned for the autumn alongside the next iPhone and Apple Watch cycle.

The most visible shift is Siri AI on the wrist. The revamped assistant is designed to handle more natural requests, continue conversations across devices, understand personal context and interact with apps more directly. On Apple Watch, that could include starting workouts, changing fitness goals, managing messages, controlling music and surfacing information through a dedicated Siri experience adapted for the smaller screen.

The new software also brings a redesigned app grid, smarter widgets, updates to Smart Stack, faster music playback, improved battery handling and expanded health and fitness tools. Workout Buddy, menstrual health notifications and language support are also being expanded, reinforcing Apple’s effort to keep the Watch positioned as both a wellness device and a personal computing accessory.

Hardware limitations are central to the decision. Apple has drawn the line around newer processors and the broader Apple Intelligence system. The Series 8 uses the S8 chip, while the first-generation Ultra and second-generation SE also fall below the new threshold. The Series 9 introduced the S9 chip, which improved on-device processing and enabled features such as the double-tap gesture. Newer Ultra and Series models build on that architecture.

The compatibility move will frustrate users who bought the first Apple Watch Ultra, launched as a premium rugged model, only to see it excluded from the next major software release four years later. The second-generation SE, marketed as a lower-cost entry point, is also affected despite being part of Apple’s current value-focused strategy until the arrival of SE 3.

The update also creates a two-tier experience among users who do receive watchOS 27. Installing the operating system will require an iPhone 11 or later, or a second-generation iPhone SE or later, running iOS 27. But Siri AI and several Apple Intelligence functions will require a paired iPhone capable of running Apple’s AI stack, starting with the iPhone 15 Pro generation and later models. That means some users with supported watches may still miss the headline features.

Apple’s decision reflects a wider industry trend as consumer technology companies tie new software features to dedicated processors, memory capacity and on-device AI performance. Smartphones, laptops and wearables are being segmented less by basic operating system compatibility and more by access to AI tools. The result is a growing gap between devices that still function well and devices that receive the most advanced software.

For Apple, the strategy could help stimulate upgrades across the Watch and iPhone ecosystem. The Watch has become a major part of the company’s wearables business, alongside AirPods and accessories, but replacement cycles have often been slower than for smartphones because many health, notification and fitness features remain useful for years. Limiting AI-led software updates to newer models may strengthen the case for replacement.

The risk is that customers may view the cut-off as abrupt, particularly because the Series 8, Ultra 1 and SE 2 remain capable devices for notifications, workouts, heart-rate tracking, sleep monitoring, Apple Pay and emergency features. Security updates may continue for older watchOS versions for a period, but users outside watchOS 27 will not receive the full set of interface changes, AI functions or future app enhancements built for the new platform.

The article Apple Watch update leaves older models behind appeared first on Arabian Post.

The controlled exercise centred on an AI agent named Pinchy, configured on the OpenClaw platform to monitor a Gmail inbox, process messages and perform tasks through connected workplace tools. The test environment used synthetic corporate material, including mock AWS IAM keys, database passwords, SSH credentials, customer records, calendar invites and internal correspondence, allowing researchers to assess the agent’s behaviour without exposing real organisations to harm.

The most serious failure came when an attacker impersonated a team lead named Dan and asked for staging-environment access during a supposed production issue. The message came from an external Gmail account, not the legitimate corporate address. Pinchy searched the inbox, found the requested access details and sent AWS IAM keys, database connection strings and SSH credentials in plaintext to the outside account.

The outcome is significant because the agent was tested under two configurations. One used generic productivity instructions, while the other added explicit safety language directing the system to watch for phishing, verify identities and avoid risky disclosures. Both configurations failed in the credential-sharing scenario, suggesting that written safety instructions alone may not reliably stop an autonomous agent when a request appears operationally urgent.

A second simulation produced a similar result with business data rather than technical secrets. A message asked the agent to send the latest customer export for a quarterly business review presentation, using the pretext of a colleague working from home. Pinchy retrieved and forwarded a CRM-style export containing 247 enterprise customer records, contact details, contract dates, customer tiers and monthly recurring revenue data worth about $1.28 million.

The tests highlight a weakness that differs from traditional prompt injection. Instead of hiding malicious commands inside web pages, documents or attachments, the attacker used a familiar workplace request delivered through a normal communication channel. The agent treated plausibility as legitimacy and prioritised completing the task over validating the sender.

Pinchy performed better in scenarios involving more obvious technical traps. When it received a fake gift-card email, the generic configuration clicked through to a phishing page and attempted to redeem the offer using fabricated credentials before later identifying the page as malicious. The stricter configuration blocked the attempt at the outset. In a separate OAuth consent test, the agent examined the redirect path of a malicious application disguised as a timesheet platform, found the destination suspicious and refused to authorise access.

That contrast points to a broader challenge for companies deploying agentic AI. These systems can inspect links, analyse domains and recognise fake login flows, yet still fail at the social judgement that employees often apply instinctively. An urgent note from a supposed colleague asking for credentials, a customer export or a payment change can exploit the agent’s core design goal: to be helpful and act quickly.

OpenClaw markets itself as a personal AI assistant that can clear inboxes, send emails, manage calendars, browse websites, use chat apps and access files or system tools depending on configuration. That breadth gives users a powerful automation layer, but it also creates a concentrated risk when the agent can read sensitive data and send messages outside the organisation.

The findings arrive as security teams are already reassessing how to govern AI agents that operate as digital workers rather than passive chatbots. Traditional email controls are built around human recipients, while identity and access systems are usually designed for applications, service accounts and employees. Autonomous agents sit between those categories: they can receive untrusted content, access private data and initiate outbound actions.

The practical risk is not limited to OpenClaw. Any agent connected to email, cloud storage, collaboration platforms, customer databases or developer tools may face the same pressure point if it can act without approval. The danger increases when agents are given broad privileges, persistent credentials or permission to communicate with first-time external recipients.

Mitigations identified through the tests are largely architectural. Agents handling inbound email should have limited access to internal systems, with permissions segmented according to the trust level of the channel that triggered the task. A message from an unverified external sender should not enable searches across customer databases, code repositories or infrastructure secrets.

The article Phishing tests expose AI inbox risks appeared first on Arabian Post.

The announcements in Taipei marked a deliberate shift in emphasis for the company, from selling individual chips to supplying integrated platforms that combine processors, accelerators, networking and software. Intel’s pitch rests on the view that AI is moving beyond the training of large models into production systems that must answer queries, manage workflows, call software tools and process data continuously inside corporate and cloud environments.

Central to the presentation was a new rack-scale AI infrastructure initiative involving Intel, SambaNova and Foxconn. The planned systems combine Intel Xeon processors with SambaNova SN-50 Reconfigurable Dataflow Units, with Foxconn providing system integration. The infrastructure is being positioned for hyperscale data centres, enterprise AI deployments and intelligence-centre workloads where inference throughput, power efficiency and predictable performance are becoming critical buying factors.

The company also highlighted a disaggregated inference model through Vector Core Compute, an enterprise inference cloud formed by Vista Equity Partners and Cambium Capital. The demonstration used Intel Xeon processors for orchestration and execution, SambaNova RDUs for decode functions and Nvidia Blackwell GPUs for prefill operations, reflecting a market trend in which AI infrastructure is being split into specialised components rather than built around a single accelerator type.

The Computex message gives Intel a clearer role in a market that has been dominated by graphics processing units. Training large AI models placed GPUs at the centre of data-centre investment, but inference workloads have different requirements. Agentic AI systems must coordinate multiple steps, retrieve data, run tools, manage memory and move information across networks. That places greater weight on central processors, interconnects and system-level optimisation.

Intel executives framed the CPU as the control plane for modern AI infrastructure. The argument is that while accelerators remain essential for model execution, CPUs are increasingly important for orchestration, concurrency, data movement and workload scheduling. As enterprises deploy AI agents into customer service, coding, research, analytics and industrial operations, the number of CPU-bound tasks around each model call is expected to rise sharply.

The company’s Xeon 6+ processor was presented as the next major data-centre product supporting that strategy. Built on Intel 18A process technology, the chip is designed for high-density, scale-out workloads and is being targeted at cloud-native, agentic AI-driven, telecom and network-intensive environments. Intel says the processor can support up to 288 Efficient-cores, 12-channel DDR5 memory, 96 lanes of PCIe Gen 5 and CXL connectivity.

Intel also promoted workload-level energy telemetry as part of the Xeon 6+ platform, giving operators more visibility into CPU energy consumption. That feature is aimed at data-centre customers under pressure to reduce power use while deploying more AI capacity. The company said the platform can help consolidate servers, reduce space requirements and support secure multi-tenant deployments through hardware-level security technologies.

Networking formed another part of the Computex package. Intel introduced its Ethernet E835 controllers and adapters for AI, cloud, edge and high-performance computing environments. The product line is designed to improve power efficiency and data movement in dense infrastructure, reflecting the growing importance of networking in systems where accelerators, CPUs, memory pools and storage must operate as coordinated platforms.

The strategy depends heavily on partners. Foxconn’s role gives Intel access to large-scale manufacturing and integration capacity for rack-level systems. SambaNova contributes AI acceleration technology aimed at enterprise inference. Intel also pointed to collaborations with Siemens, Hitachi, Echo Neurotechnologies and Greenstone Biosciences, signalling a push into industry-specific AI applications rather than a purely horizontal data-centre offering.

The company used Computex to extend the message beyond the data centre. It said more than 130 companies are adopting or testing Intel Core Ultra Series 3 processors for edge devices, robotics and physical AI systems. The platform integrates CPU, GPU and neural processing capability in a single system-on-chip, targeting use cases in manufacturing, healthcare, hospitality, education and automation. Intel also introduced OpenVINO Physical AI as an open-source robotics framework for developers.

The timing is important for Intel as it seeks to regain momentum after years of manufacturing delays, market-share pressure from AMD and the rapid expansion of Nvidia’s AI ecosystem. The company is trying to reposition itself as a systems supplier for the next phase of AI deployment, where enterprise customers may require more balanced combinations of CPUs, accelerators, networking and software rather than GPU-heavy clusters alone.

The article Intel pushes AI systems beyond the chip appeared first on Arabian Post.

The article Anthropic opens Fable model access appeared first on Arabian Post.

The shift marks a technical break from conventional mesh-based photogrammetry, the approach that has long powered digital city models but often produced warped tree canopies, distorted cables, softened building edges and streetscapes that appeared convincing only from a distance. Gaussian splatting uses dense collections of semi-transparent 3D points, or “splats”, to reproduce shape, colour, light and depth with higher visual fidelity, allowing complex details to hold together from multiple viewing angles.

For Apple Maps, the change could strengthen one of the platform’s most visually distinctive features: immersive 3D city navigation. The company has spent years expanding detailed city experiences, Look Around street-level views and Flyover-style aerial perspectives across major urban centres. A move towards splat-based rendering would give the service a more photorealistic layer, especially in areas where traditional reconstruction struggles with fine structures such as trees, overhead wires, roof edges, balconies, traffic signs and irregular street furniture.

The scenes are understood to be generated from oblique aerial imagery, captured at angles rather than directly from above. That matters because angled image sets provide more information about façades, street corridors, rooflines and vertical surfaces. When processed through Gaussian splatting methods, those images can produce scenes that preserve ground-level detail without relying entirely on the rigid triangle meshes used in older 3D mapping pipelines.

The technology has gained rapid traction across computer vision, geospatial mapping, digital twins and spatial computing because it can render photorealistic 3D environments in real time. Instead of converting imagery into a conventional polygon model, Gaussian splatting represents a scene through millions of small, soft volumetric elements. Each carries data about position, scale, orientation, colour and transparency, allowing a viewer to move through a scene while the system blends these elements into a coherent image.

Apple’s interest fits a wider strategy across mapping, spatial media and Vision Pro. Its software ecosystem is increasingly built around immersive visual experiences, with developers encouraged to create richer 3D environments for apps, websites and spatial computing workflows. Better Maps data also supports navigation, tourism, property search, gaming, urban planning and augmented reality applications, where realistic geometry and visual stability are essential.

The most immediate consumer impact will be visual quality. Older photogrammetry often looked impressive at city scale but broke down on inspection. Trees could turn into indistinct green clumps, bridges could smear at their edges, and thin infrastructure could appear bent or melted. Gaussian splatting is better suited to preserving such visual complexity, particularly when source imagery has strong multi-angle coverage. For users, that could mean smoother city previews, more recognisable landmarks, cleaner neighbourhood views and a stronger sense of place before arriving at a destination.

The competitive implications are also significant. Google Maps and Google Earth remain the global benchmark in mapping coverage, data volume and street-level imagery. Apple, however, has been investing steadily in higher-quality cartography, privacy-focused navigation and curated city experiences. A sharper 3D layer would not erase Google’s scale advantage, but it could help Apple compete on visual quality in supported locations and deepen the appeal of Maps across iPhone, iPad, Mac and Vision Pro.

The rollout is likely to be selective at first. High-quality splat-based city models require dense, well-aligned imagery, careful processing and efficient compression for delivery to consumer devices. Dense urban cores, landmarks, tourism districts and cities already covered by Apple’s detailed 3D experiences are the most likely early candidates. Broader coverage would depend on aerial capture cycles, licensing arrangements, local regulation and the cost of processing vast quantities of imagery.

There are also unresolved questions. Photorealistic city models raise privacy and security sensitivities, particularly around sensitive facilities, residential areas and real-time accuracy. Mapping companies typically blur faces and licence plates in street-level imagery, while aerial and 3D datasets may require additional safeguards. More realistic scenes could also increase pressure on platforms to keep visual data current, because outdated building layouts or construction zones may become more noticeable when rendered with greater fidelity.

The article Apple Maps sharpens 3D city views appeared first on Arabian Post.

The new Siri AI is available immediately for developer testing across iOS 27, iPadOS 27, macOS 27 and visionOS 27, with support for watchOS 27 due in a later developer beta. A wider beta for users is scheduled later this year, initially for supported devices set to English, before a staged expansion to more languages.

The announcement marks Apple’s most direct attempt to close the gap with rivals that moved faster in generative AI. Siri AI has been rebuilt to handle more conversational exchanges, understand what is on a user’s screen, draw on personal context from apps such as Messages, Mail and Photos, and search the web for answers. Apple is also giving Siri a dedicated app, allowing users to revisit past conversations across devices through iCloud syncing.

Craig Federighi, Apple’s senior vice-president of Software Engineering, said the upgraded assistant was designed to help users find information and complete tasks through deeper app integration. The system can draft emails, edit and share photos, answer questions about visible content, and respond to follow-up queries in a more natural exchange than the older voice assistant.

Apple’s push is built around its Apple Intelligence platform, with processing split between on-device models and Private Cloud Compute for heavier tasks. The company has emphasised privacy as a central difference from competing AI systems, saying personal data used in cloud-based requests is not stored or made available to Apple. That claim will remain under scrutiny as Siri AI gains broader access to messages, emails, images and screen content.

Availability will be uneven at launch. Siri AI and the wider Apple Intelligence upgrades will work on iPhone 16 models or later, iPhone 15 Pro and iPhone 15 Pro Max, iPad mini with A17 Pro, iPads with M1 or later, Macs with M1 or later, Apple Vision Pro, Apple Watch Series 9 or later, Apple Watch Ultra 2 or later, and Apple Watch SE 3 when paired with a compatible iPhone. Some of the most powerful on-device AI functions will require newer hardware with higher memory capacity.

Apple said Mac, Apple Watch and Vision Pro users in the European Union will be able to access Siri AI when using a supported language, but the feature will not initially be available on iOS 27 and iPadOS 27 in the bloc. The company has linked the delay to regulatory requirements under the Digital Markets Act. Siri AI and other new Apple Intelligence tools will also not be available in China at the outset as Apple works through regulatory approvals.

Alongside Siri, Apple announced a redesigned set of parental controls aimed at giving families tighter command over children’s device use. The tools include a simpler child account setup, a recommended set of essential apps, Ask to Browse for Safari, Time Allowances for app categories, and an updated Screen Time dashboard.

Child accounts will allow age-based protections such as limiting adult websites, restricting media by age rating and setting App Store boundaries. Parents will be able to choose which apps a child can use from the start, approve new websites in Safari, and manage communications across Messages, FaceTime and Phone. Communication Safety, already used to blur nudity in Messages and FaceTime for under-18 users, will expand to intervene when gore or violent images and videos are detected.

Time Allowances will let parents set limits across app categories including entertainment, games and social media. Apple is also adding daily schedules so families can restrict access during school hours, meals or other periods. Screen Time will show average usage and frequently used apps more clearly, while giving parents faster controls to pause access or extend time when needed.

Developers are being given additional tools to support age-appropriate experiences inside apps. These include SensitiveContentAnalysis, PermissionKit and a Declared Age Range API that lets apps adapt content without sharing a child’s exact birth date.

The article Apple sharpens Siri with safer AI appeared first on Arabian Post.

Eligible users on Free, Go, Plus and Pro plans, along with self-serve ChatGPT Business accounts, are being given access to the feature through the Security section of ChatGPT settings. The rollout marks a notable shift from its initial positioning for higher-risk enterprise users, including executives, security teams and organisations handling sensitive material.

Lockdown Mode is designed to reduce the risk that malicious instructions hidden in web pages, files or connected services could cause ChatGPT to send sensitive information outside the conversation. The control does not claim to block every prompt injection attack, but it limits the final route through which stolen or exposed data might leave the system by restricting outbound network activity.

When enabled, the mode curbs or disables several ChatGPT capabilities that depend on live links to the internet or external services. Live web browsing is limited to cached content, meaning search results may be incomplete, unavailable or outdated. Deep Research is disabled, Agent Mode is turned off, Canvas-generated code cannot be approved for network access, and ChatGPT cannot download files for data analysis. Users can still upload files manually, and image generation remains available where the account otherwise supports it.

The change reflects a broader recalibration across the AI industry as assistants move from answering isolated prompts to handling tasks involving email, documents, websites, shopping, coding tools, workplace apps and external databases. These connections make systems more useful, but they also widen the attack surface. Prompt injection remains one of the most difficult security problems in AI because the harmful instruction may appear as ordinary content inside a page, document or tool output.

OpenAI’s approach is to trade some convenience for tighter control. A user working with sensitive internal documents, legal material, financial data, confidential reporting notes or business records may choose Lockdown Mode when the risk of external leakage outweighs the value of live web access or automated agent workflows. For ordinary queries, the setting may feel restrictive because several high-value ChatGPT tools become unavailable or limited.

The company has also positioned the feature as part of a layered defence system, rather than a standalone cure. Existing protections include sandboxing, restrictions against URL-based exfiltration, monitoring, enforcement mechanisms, enterprise controls, audit logs and role-based permissions. Lockdown Mode adds a more conservative operating state for users who want stronger boundaries around external interaction.

For personal and self-serve Business accounts, the feature can be turned on from Settings under Security. Once active, a status message appears above the composer. Users may disable it for an individual chat, allowing some flexibility when they need a less restricted session. Lockdown Mode and Developer Mode cannot run at the same time; switching on one turns off the other.

Managed workspaces have a different route. Administrators can use role-based access controls to create a custom Lockdown Mode role and assign it to members or groups. This allows organisations to apply stricter controls to selected employees rather than imposing the setting across an entire workspace. Administrators are also expected to review app permissions, connector access and write actions because app access inside ChatGPT does not override the permissions already set in the connected source system.

The connector rules are especially important for business users. For personal accounts and self-serve Business accounts, Lockdown Mode allows connectors that rely on synced data while blocking live connector access and connector write actions. Some connected experiences, including finance and shopping-agent functions, may be unavailable. In managed workspaces, apps, model context protocols and connectors depend on workspace settings and custom roles, so the restrictions may vary by organisation.

OpenAI has made clear that Lockdown Mode does not change every privacy or data setting. It does not alter memory, file uploads, conversation sharing, or whether chats may be used to improve models. Those controls remain separate and depend on the user’s account and workspace settings. It also does not affect Codex network access, which is governed by its own settings and elevated-risk warnings.

The wider release comes as AI providers face growing pressure to make security controls understandable to non-specialist users. Prompt injection can be difficult to explain because it does not always resemble a conventional hack. A hostile instruction may be embedded in a webpage, pasted text, email, document or third-party app output, then interpreted by the AI system as something to follow. The danger rises when an assistant has access to private context and tools capable of sending data elsewhere.

The article ChatGPT lockdown mode reaches wider user base appeared first on Arabian Post.

The finding has sharpened concern across the software supply chain because the CRA’s vulnerability and incident reporting duties begin on 11 September 2026, well before the wider compliance regime takes full effect on 11 December 2027. The law, formally Regulation 2024/2847, entered into force on 10 December 2024 and applies to products with digital elements placed on the EU market, including software, connected hardware and certain standalone components.

The Open Source Security Foundation has said 66% of open source practitioners are either unaware of the CRA deadline or not prepared for it, despite the law already being in force. Its wider research points to uncertainty among maintainers, manufacturers and open-source stewards over who carries responsibility for reporting exploited vulnerabilities, maintaining security processes and documenting compliance across projects that often rely on unpaid contributors.

The compliance challenge is particularly acute because modern software products depend heavily on open-source components. Enterprise platforms, cloud services, mobile applications, industrial systems and internet-connected devices frequently include packages maintained by distributed communities outside the commercial structures that ultimately place products on the EU market. That separation between upstream code creation and downstream commercial use has become one of the central tensions in the CRA debate.

The regulation treats open source differently depending on whether it is offered commercially. Free and open-source software that is not monetised and is not made available on the market in the course of commercial activity is generally outside the main manufacturer obligations. Individual developers contributing code to projects that are not under their responsibility are also not treated as manufacturers. However, companies that place products containing such software on the EU market remain responsible for compliance, while a new category of open-source software steward covers legal entities that provide sustained support for projects intended for commercial use.

Open-source software stewards face a lighter regime than manufacturers, but they still have obligations. These include maintaining a cybersecurity policy, supporting secure development, handling vulnerabilities and cooperating with market surveillance authorities. They must also report actively exploited vulnerabilities and severe security incidents affecting relevant products, although the CRA does not subject stewards to administrative fines for infringements.

Manufacturers face a tougher framework. Products covered by the Act must be designed, developed and maintained with cybersecurity in mind throughout their lifecycle. Companies will need processes for vulnerability handling, software updates, technical documentation, conformity assessment and incident reporting. Serious breaches can lead to fines of up to €15 million or 2.5% of global annual turnover, whichever is higher, while other infringements may attract lower but still significant penalties.

The first major test arrives with the September 2026 reporting duty. Manufacturers will need to report actively exploited vulnerabilities and severe incidents through the EU reporting architecture, involving ENISA and national computer security incident response teams. For companies that ship products with long chains of open-source dependencies, this means identifying which components are present, whether they are maintained, how vulnerabilities are tracked and who can act quickly when exploitation is detected.

That requirement has pushed software bills of materials, vulnerability disclosure policies, secure build systems and dependency mapping higher on boardroom agendas. Tools such as OpenSSF Scorecard, SLSA and project security baselines are gaining attention as organisations seek practical ways to measure upstream risk and demonstrate due diligence. Larger technology companies including Red Hat, Microsoft, GitHub and Ericsson have been active in policy and standards discussions, while foundations and working groups are trying to translate legal obligations into workflows that fit open collaboration.

Smaller developers and SMEs remain a weak point. Many lack legal teams, security staff or dedicated compliance budgets, even when their software is embedded in commercial products sold across Europe. OpenSSF has warned that weak readiness among smaller participants could reduce project diversity, increase pressure on volunteer maintainers and shift costs towards communities that were not designed to operate as regulated suppliers.

The CRA was created after a series of software supply-chain incidents exposed the fragility of widely used digital infrastructure. The Log4j vulnerability, attacks on package repositories and repeated exploitation of outdated dependencies strengthened the case for mandatory security-by-design rules. The EU’s approach seeks to make manufacturers responsible not only for product functionality at release, but also for security support and vulnerability management after deployment.

The article Open source faces CRA awareness gap appeared first on Arabian Post.

The article Capital.com brings AI tools into trading accounts appeared first on Arabian Post.