The First AI Deepfake Detection System Launches in Dubai

Dubai will launch an AI model for identifying deepfake videos by the end of this year; a model that can not only determine the falseness of a video but also identify the manipulated parts.

The Dubai Electronic Security Center (DESC) unveiled this model named "SARAAB" on the first day of the three-day GISEC Global Cybersecurity Conference and Exhibition.

The SARAAB model analyzes the video and uses a "heat map" to show which parts of the face have been manipulated. Even if only a portion of a long fake video is altered, the model can identify that specific part.

According to Hassan Majid Al-Khazraji, Chief AI Officer at DESC, if a video is two minutes long, the entire two minutes will be analyzed; therefore, if, for example, only the last 30 seconds have been manipulated, that part will also be identified by the model.

This model is set to be launched by the end of 2026, and according to DESC, its accuracy will reach 91%; a figure that is among the highest reported globally for such models.

The SARAAB model is expected to be released on the open-source platform Hugging Face so that users can view its code and modify or develop it if needed. DESC states that this is the first AI model developed by a government entity in the Arab region.

The center has been working on the development of SARAAB for about six months. The idea for its creation arose from a specific concern: the use of deepfake technology by malicious individuals to create fake advertisements using the faces of individuals and organizations.
Source: Gulf Times