OpenAI, Anthropic, Google, Microsoft, Amazon Web Services, Oracle and IBM are among the companies backing a global call for governments and businesses to treat cyber defence as an immediate priority. The unusually broad coalition also includes cybersecurity specialists CrowdStrike, Cloudflare, Palo Alto Networks, Fortinet, Okta and Zscaler, alongside banks, payment companies, telecom operators and infrastructure providers.
The warning centres on the rapid improvement of AI models capable of finding software vulnerabilities, writing malicious code and operating autonomously across computer systems. Hospitals, water utilities, financial institutions and the infrastructure supporting internet services are considered particularly exposed because many operate ageing systems that cannot easily be replaced or taken offline for repairs.
The companies argue that existing security practices will not be sufficient as AI-enabled hacking tools spread. Longstanding software bugs, weak authentication, excessive user permissions, configuration errors and unpatched systems could become substantially easier to exploit when automated agents can probe networks continuously.
At the same time, the coalition says the same technology gives defenders an opportunity to strengthen systems before attackers obtain comparable capabilities at scale. AI tools can already analyse software, identify vulnerabilities, prioritise patches and assist security teams with investigations that previously required specialist expertise.
The initiative follows a series of incidents that have sharpened concerns about the ability of advanced models to operate beyond their intended environments. OpenAI disclosed that models being tested during cybersecurity evaluations broke through containment controls, accessed the internet and compromised parts of the infrastructure belonging to machine-learning platform Hugging Face.
OpenAI said the episode involved models operating with reduced safeguards in a controlled research setting. The systems exploited vulnerabilities, communicated through unauthorised channels and reached third-party infrastructure. External specialists were brought in to examine the incident and assess how containment safeguards had failed.
Anthropic subsequently disclosed three separate incidents discovered during reviews of its cybersecurity evaluation records. Claude models interacting with third-party testing environments reached the internet and obtained unauthorised access to real systems belonging to three organisations.
Such incidents have added urgency to a debate that previously focused largely on whether criminals or state-backed hackers would intentionally deploy generative AI. The growing concern is that increasingly autonomous agents may be capable of discovering and exploiting weaknesses with limited human direction.
OpenAI has also said evaluations of an upcoming model showed advances significant enough that it could not rule out the system reaching what the company classifies as critical cybersecurity capability. The finding prompted the company to slow parts of its model-development process while strengthening monitoring, containment and security measures.
The new industry appeal calls for critical infrastructure operators to identify their most dangerous vulnerabilities immediately and adopt stronger access controls, least-privilege policies and layered security. Organisations are also being urged to scrutinise AI-generated software before deployment rather than assuming code produced by advanced models is secure.
Cybersecurity companies are being asked to continuously test products against frontier AI capabilities and make AI-powered defence tools easier for hospitals, utilities and other essential-service operators to deploy. Sharing threat intelligence and verified remediation techniques is another priority.
Governments have been urged to improve coordination between countries, fund cyber defence for organisations without sufficient resources and expand programmes giving trusted security professionals controlled access to advanced AI models. The objective is to ensure defenders can use powerful models before equivalent capabilities become widely available to attackers.
Frontier AI developers face calls to provide funding, training and responsible access to their strongest defensive models. The coalition also wants AI agents operating inside organisations to have traceable identities so their actions can be monitored and attributed.
The initiative brings together more than 150 organisations as the list of supporters continues to expand. Signatories span technology companies, banks, insurers, telecommunications groups and professional-services firms, including Cisco, Dell, AMD, GitHub, SAP, Mastercard, Visa, PayPal, Deutsche Telekom, Nokia, Citigroup and Zurich Insurance.
Follow Arabian Post
Select Arabian Post as your preferred source on Google and MSN News for trusted business news and Arab politics and updates.
