Dubai is Building an Open-Source AI Model to Detect Deepfake Videos

Dubai is set to unveil SARAAB, an open-source artificial intelligence model created to identify deepfake videos by examining facial features and motion patterns. The Dubai Electronic Security Centre (DESC) states that SARAAB identifies concealed synthetic layers in video footage, even when only part of the clip has been tampered with.

The model is slated for release by the end of 2026, as announced during GISEC Global 2026, an event held from September 16 to 18 at the Dubai Exhibition Centre in Expo City Dubai. The revelation was initially reported by Khaleej Times, while Gulf News stated that an Emirati team within DESC developed the model.

SARAAB operates by analyzing facial features via a heat map to spot indications of artificial manipulation. According to Hassan Majid Alkhazraji, a Senior Artificial Intelligence Executive at DESC, the system can assess an entire video, even if only a small segment has been altered. For instance, if the manipulation occurs solely in the final 30 seconds of a two-minute video, SARAAB would still scrutinize the complete two-minute clip.

The organization claims that SARAAB can achieve 91% accuracy, although the announcement does not disclose the dataset, evaluation method, or testing conditions used to attain this percentage. Consequently, the 91% figure should be regarded as a claim from DESC rather than an independently validated benchmark.

DESC intends to release SARAAB through developer platforms, enabling AI engineers, cybersecurity specialists, and researchers to assess and enhance the system. While this open-source approach could empower organizations to utilize the model for detecting dubious video content, it's crucial to note that relying on a single detection model cannot conclusively confirm a video's authenticity, especially as synthetic video techniques continue to evolve.