Dubai’s AI Deepfake Detector Can Identify Manipulated Videos in 7 Seconds

Dubai’s Electronic Security Center (DESC) has developed an open-source AI model capable of detecting deepfake videos using clips as short as seven seconds, giving investigators another tool for examining manipulated digital content.

Named SARAAB, the model achieved up to 91% accuracy in DESC’s testing and recorded 89% accuracy when evaluated against previously unseen data. DESC said the model also performed better than the other open-source deepfake detection models included in its comparison.

The seven-second capability is particularly relevant to forensic investigations involving short-form videos, where investigators may have only a limited recording available for analysis. Other open-source models evaluated by DESC generally required videos of at least 15 seconds for effective analysis.

SARAAB was developed entirely by an Emirati team at DESC and was officially launched in September during GISEC Global 2026. DESC plans to release the model through GitHub and Hugging Face, allowing researchers, cybersecurity professionals and developers to test and build on the technology.

For digital-forensics teams, tools such as SARAAB could become increasingly important as investigators face growing volumes of AI-generated and manipulated video evidence.