Arabian Post Staff -Dubai
The open-source model was developed entirely by a team at the Dubai Electronic Security Centre, or DESC, and is designed to analyse video frame by frame for signs of synthetic alteration. The centre says SARAAB is the first deepfake-detection model of its kind developed by a government entity in the Arab region.
DESC plans to make the model available through developer platforms by the end of 2026, allowing artificial intelligence engineers, cybersecurity specialists and researchers to test the system, examine its code, build on its capabilities and contribute to further development.
Hassan Majid Alkhazraji, senior artificial intelligence executive at DESC and one of the project’s developers, said the system examines an entire video rather than sampling only selected portions. That approach is intended to identify manipulated segments that may appear only briefly within otherwise authentic footage.
The model uses a heat map to highlight facial areas where manipulation may have occurred. Alkhazraji said a two-minute video would be analysed across its full duration, including any section in which deepfake content appeared only near the end. DESC says the current model can achieve detection accuracy of up to 91 per cent.
The centre has not presented the accuracy figure as a guarantee across every type of manipulated video. Deepfake-detection performance can vary depending on compression, editing techniques, image quality, the generation method used and whether material has been altered after creation, making continued testing important as synthetic-media tools evolve.
SARAAB was introduced on the opening day of GISEC Global, held from September 16 to 18 at Dubai Exhibition Centre in Expo City Dubai. The cybersecurity event brought together government bodies, technology companies, security specialists and researchers focused on threats ranging from AI-enabled fraud to infrastructure protection.
Amer Sharaf, chief executive of DESC’s Cyber Security Systems and Services Sector, said artificial intelligence was accelerating established forms of cybercrime by making fake identities, cloned voices and manipulated videos easier to produce. He urged users to be especially cautious when apparently familiar people make unusual requests, particularly demands involving money or sensitive information.
Sharaf said SARAAB scans individual frames and applies a facial heat map to identify areas that may have been synthetically altered. The system is intended to give investigators and users an additional technical method for evaluating content that can be difficult to assess visually.
The deepfake initiative formed part of a wider package of cybersecurity measures announced by DESC at GISEC. The centre also introduced the Dubai Cyber Skills Framework, which is intended to provide a common reference for qualifications, professional certifications, training routes and career progression across the cybersecurity sector.
A third initiative, the ISR Auditor Certification Programme, is aimed at training auditors and information-security professionals to assess compliance with Dubai’s Information Security Regulation using consistent standards and assessment methods.
Yousuf Hamad Al Shaibani, chief executive of DESC, said the initiatives reflected Dubai’s effort to anticipate shifts in cyber risk while building specialist expertise locally. He said the centre wanted solutions developed in Dubai to have practical relevance beyond the emirate as threats increasingly cross organisational and national boundaries.
DESC’s decision to release SARAAB as open source is intended to widen scrutiny and participation around the model. Researchers will be able to examine its performance, test it against new deepfake-generation techniques and propose improvements as attackers adopt more sophisticated tools before it is released publicly later this year.
The focus on video manipulation comes as generative AI systems make convincing synthetic content faster and cheaper to produce. Security officials have warned that deepfakes can be used in impersonation scams, fraudulent advertising, social-engineering attacks and attempts to mislead audiences by fabricating statements or actions.
Follow Arabian Post
Select Arabian Post as your preferred source on Google and MSN News for trusted business news and Arab politics and updates.
