Imagine watching a video of a famous person making a shocking statement.

The face looks real.

The voice sounds real.

The expressions look natural.

And the video is convincing enough that you immediately believe it.

But there is one problem.

The person never actually said it.

This is the growing problem of deepfake technology, where artificial intelligence can manipulate or generate video content that looks remarkably authentic.

Now Dubai is developing a new AI system designed specifically to fight this problem.

The Dubai Electronic Security Centre, or DESC, has unveiled SARAAB, an artificial intelligence model designed to detect deepfake videos. The model was announced at GISEC Global 2026 in Dubai and is being developed as an open-source tool that researchers, developers and cybersecurity experts can test and improve.

And what makes SARAAB particularly interesting is that it is designed to analyze an entire video, rather than simply checking one image or one frame.

What Is SARAAB?

SARAAB is an AI-powered deepfake detection model developed by an Emirati team at Dubai’s Electronic Security Centre.

According to DESC, it is the first model of its kind developed by a government entity in the Arab region. The centre says the model was created to respond to the growing threat posed by AI-generated and manipulated video content.

The basic idea is simple.

AI can now be used to create fake videos.

So another AI system can be used to look for the digital fingerprints left behind by that manipulation.

SARAAB analyzes facial features and uses a heat-map approach to identify areas that may contain synthetic or manipulated elements.

Instead of simply saying:

“This video is fake.”

The system is designed to help identify where the manipulation may have occurred.

That could make the technology more useful for investigators, cybersecurity teams, media organizations and researchers.

Why Deepfakes Are Becoming a Serious Problem

Deepfakes are not a completely new technology.

But generative AI has made the creation of convincing synthetic media much easier.

A manipulated video can potentially be used to make someone appear to say something they never said.

That creates problems far beyond social media entertainment.

Imagine a fake video of a company CEO announcing a major decision.

Or a fake government official making an emergency announcement.

Or a fake celebrity promoting a financial product.

Or a fake person appearing in a video call.

The danger isn’t simply that people can create fake content.

The bigger problem is that people may believe it.

Once realistic synthetic video becomes difficult to distinguish from authentic footage, trust itself becomes a cybersecurity issue.

This is the environment in which SARAAB is being introduced.

How Does SARAAB Detect a Deepfake?

One of the key features described by DESC is SARAAB’s use of a heat map.

The system examines facial features and looks for signs of synthetic manipulation.

The resulting visualization can highlight areas where the model detects potential manipulation.

Think of it like an AI investigator pointing at a person’s face and saying:

“Look here. Something about this part of the video may have been artificially generated or altered.”

This approach can be particularly useful because deepfakes do not necessarily manipulate an entire video.

Sometimes only a small section may be changed.

A person could appear normally for most of a video and then have their face manipulated for a few seconds.

That’s where SARAAB’s full-video analysis becomes important.

It Doesn’t Just Check One Frame

According to DESC representatives quoted by Khaleej Times, SARAAB can analyze an entire video.

For example, imagine a two-minute video.

The first 90 seconds could be completely authentic.

Only the final 30 seconds might contain a deepfake.

The system is designed to analyze the whole two minutes rather than stopping after finding apparently normal footage.

This matters because checking only selected frames could potentially miss manipulation that appears later.

A full-video approach gives the detection system a broader view of what is happening throughout the footage.

That doesn’t mean every deepfake will automatically be detected.

But it shows that SARAAB is being designed around a more comprehensive analysis process.

Dubai Says SARAAB Has 91% Accuracy

One of the biggest claims surrounding the new system is its reported accuracy.

DESC has said that SARAAB has an accuracy rate of 91 percent, which the centre described as among the highest in the world.

However, this number should be understood carefully.

An accuracy percentage by itself doesn’t tell us everything about a detection model.

To properly evaluate a system like this, researchers would want to know details such as:

What dataset was used?

How many real videos were tested?

How many manipulated videos were tested?

What types of deepfakes were included?

How does the model perform against new manipulation techniques?

And how does it compare with other detection systems under the same conditions?

The current public reporting does not provide enough information to independently verify the 91% figure across all these conditions.

So for now, 91% should be treated as DESC’s reported performance claim, rather than as a universally established benchmark.

That distinction is important because deepfake technology itself is constantly changing.

Why Open Source Could Be Important

Perhaps one of the most interesting decisions by Dubai is that SARAAB is planned as an open-source AI model.

DESC says the model will be made available through developer platforms so AI engineers, cybersecurity professionals and researchers can test it, build on it and contribute to its development.

Khaleej Times reported that SARAAB is plan