DUBAI: Dubai has developed an artificial intelligence model designed to identify deepfake videos, giving cybersecurity researchers a locally built tool capable of analysing footage as short as seven seconds.
Named SARAAB, the model was developed entirely by an Emirati team at the Dubai Electronic Security Centre (DESC). The centre describes it as the first open-source AI model for deepfake detection developed by a government entity in the Arab region.
DESC unveiled SARAAB during GISEC Global 2026 at the Dubai Exhibition Centre in Expo City Dubai on September 16. The model is intended to help address the growing difficulty of distinguishing genuine digital content from video altered or generated using AI.
According to figures reported by DESC, SARAAB has reached up to 91% accuracy in detecting deepfakes. A subsequent report said the model recorded 89% accuracy when tested against data it had not previously encountered during training. The 91% figure is therefore best presented as a reported result from DESC’s testing rather than as an independently established industry benchmark.
Why seven-second videos matter
One of SARAAB’s notable capabilities is its ability to analyse videos as short as seven seconds.
That is particularly relevant to the way people now consume and share information online, where short videos can quickly circulate across social media platforms. According to the reported comparison, other open-source models assessed by DESC required clips of at least 15 seconds for analysis.
The model is also designed to examine video rather than simply treating a single frame as evidence. Reporting from GISEC said SARAAB can use a heat map to highlight facial areas where manipulation is detected, helping investigators see which parts of an image may have been altered.
That approach could be useful when a video contains a mixture of authentic and manipulated material. DESC officials have explained that the system can examine an entire video, meaning a manipulated section appearing only toward the end of otherwise genuine footage can still be assessed.
Open-source model aimed at researchers
DESC’s decision to make SARAAB open source is another important part of the project.
The centre says the model will be made available through developer platforms, allowing AI engineers, cybersecurity specialists and researchers to test it, build applications around it and contribute to further development.
The approach also reflects the changing nature of the deepfake problem. Detection systems must keep evolving as generative AI tools become better at producing convincing faces, voices and video sequences.
For cybersecurity teams, an open model offers an opportunity to test performance against new forms of synthetic media rather than relying solely on a closed commercial detection system.
Dubai’s wider cybersecurity push
SARAAB was one of three cybersecurity initiatives unveiled by DESC at GISEC Global 2026.
The centre also introduced the Dubai Cyber Skills Framework, designed to provide a common reference for cybersecurity qualifications, certifications, training and career development. It additionally announced an ISR Auditor Certification Programme aimed at improving the consistency of information-security compliance auditing.
The initiatives form part of Dubai’s wider effort to develop local cybersecurity capabilities as digital threats become more sophisticated.
The deepfake challenge extends beyond misinformation. AI-generated or manipulated video can potentially be used for impersonation, fraud, reputational attacks and the creation of convincing false evidence. Dubai cybersecurity officials have also warned that AI is making traditional signs of online fraud less reliable.
For UAE users, that means a convincing video should not automatically be treated as proof of an event, statement or person’s actions. Verification through trusted sources remains important, particularly when digital content is being used to support financial requests, urgent instructions or claims involving public figures and organisations.
What happens next?
DESC plans to make SARAAB available to the wider technical community through developer platforms, opening the system to additional testing and development.
The project does not mean every AI-generated video can automatically be classified as genuine or fake. Detection accuracy varies according to the technology used to create manipulated content, the material being analysed and the conditions under which a model is tested.
What SARAAB represents is a move toward developing local, openly accessible tools for digital-content verification — an area likely to become increasingly important as synthetic media becomes harder for people to distinguish from authentic footage.
What is SARAAB?
SARAAB is an open-source artificial intelligence model developed by the Dubai Electronic Security Centre to detect deepfake videos. DESC says it was developed entirely by an Emirati team.
How accurate is Dubai’s SARAAB AI model?
DESC has reported accuracy of up to 91% in its testing. A separate test cited by Emirates 24|7 reported 89% accuracy on previously unseen data.
How short a video can SARAAB analyse?
According to the reported DESC testing, SARAAB can analyse videos as short as seven seconds.
How does SARAAB identify deepfakes?
The system analyses video footage and can use heat-map visualisation to identify facial areas showing signs of manipulation.
Who developed SARAAB?
SARAAB was developed by an Emirati team at the Dubai Electronic Security Centre (DESC).
Is SARAAB open source?
Yes. DESC says SARAAB will be made available through developer platforms so researchers, AI engineers and cybersecurity specialists can test and contribute to its development.
When was SARAAB unveiled?
DESC unveiled SARAAB on September 16, 2026, during GISEC Global 2026 at the Dubai Exhibition Centre in Expo City Dubai.
