With SARAAB AI built to detect deepfakes at 91% accuracy, Dubai has redefined open-source cybersecurity. Unveiled by the Dubai Electronic Security Centre (DESC) at the GISEC Global 2026 exhibition held at Expo City Dubai from September 16 to 18, 2026 this state-of-the-art model targets synthetic media manipulation head-on.
Synthetic media is evolving at a terrifying pace. Distinguishing real footage from manipulated clips is now a critical challenge for governments, enterprises, and media outlets globally. Developed entirely by a specialized team of Emirati researchers and engineers, SARAAB represents a massive leap forward for Dubai’s Cyber Security Strategy.
By hosting this model on Hugging Face in late 2026, Dubai aims to democratize deepfake detection. This move allows developers and security researchers worldwide to collaborate on refining the technology.
Quick Summary
- What is SARAAB AI? It is an open-source AI model engineered specifically to identify facial and motion manipulation in videos.
- Who developed SARAAB AI? The model was built entirely by a dedicated team of Emirati cybersecurity and AI engineers at the Dubai Electronic Security Centre (DESC).
- How does SARAAB AI detect deepfakes? It performs frame-by-frame analysis using heat map technology to flag even minor, short-duration alterations.
- When will SARAAB AI be available? It is scheduled for public release on the Hugging Face platform by late 2026.
- What is the current status of the project? The technology was officially announced and demonstrated at GISEC Global 2026 in mid-September.
Technical Specifications: SARAAB AI Model
| Feature / Specification | Details |
|---|---|
| Developer | Dubai Electronic Security Centre (DESC) |
| Development Team | 100% Emirati Cybersecurity & AI Engineers |
| Model Type | Open-source Deepfake Video Detection |
| Platform Host | Hugging Face (Expected late 2026) |
| Target Launch Date | Late 2026 |
| Claimed Accuracy | 91% |
| Primary Methodology | Heat Map Generation, Facial Landmark Analysis, Motion Tracking |
| Analysis Depth | Full-length, frame-by-frame scanning (detects partial edits) |
| Target Users | Cybersecurity Researchers, AI Developers, Media Verification Teams, Enterprises |
How SARAAB Works: Under the Hood
Unlike basic detection tools that scan a video as a single file, SARAAB uses a rigorous, multi-layered approach to identify synthetic manipulation.
“`
[Input Video File]
│
▼
[Frame-by-Frame Decomposition] (Scans every millisecond)
│
▼
[Facial Landmark & Motion Tracking] (Analyzes micro-expressions & movement)
│
▼
[Heat Map Generation] (Highlights anomalies and synthetic layers)
│
▼
[Final Classification] (Provides confidence score up to 91% accuracy)
“`
1. Frame-by-Frame Decomposition
Deepfake creators often alter only specific parts of a video to save rendering time. A two-minute clip might only contain ten seconds of manipulated footage.
SARAAB counters this by breaking down the entire video file and analyzing every single frame. Even if a manipulation lasts for less than a second, the model is designed to flag it instantly.
2. Facial Feature and Motion Tracking
The model tracks subtle facial muscle movements, eye blinking patterns, and blood flow indicators. These micro-expressions often look unnatural in AI-generated videos.
It actively looks for inconsistencies between the movement of the eyes, mouth, and head.
3. Heat Map Generation
When SARAAB detects a potential modification, it generates a visual heat map over the video frame.
This heat map highlights the exact coordinates where synthetic layers have been applied. It gives security analysts clear visual proof of where the video was tampered with.
“SARAAB does not just give a yes-or-no answer; it dissects the video. If only the last 30 seconds of a two-minute clip are manipulated, our model identifies exactly where the edit occurred using advanced heat map visualization.”
: Hassan Majid Alkhazraji, Senior AI Executive at DESC, speaking at GISEC Global 2026.
SARAAB vs. Existing Deepfake Detectors
To understand SARAAB’s place in the global security sector, it helps to compare it with existing proprietary and open-source detection tools available in 2026:
| Feature | SARAAB (DESC) | FakeCatcher (Intel) | Video Authenticator (Microsoft) |
|---|---|---|---|
| Availability | Open-source (Hugging Face) | Proprietary / Commercial | Proprietary / Enterprise-only |
| Primary Metric | Heat Maps & Motion Anomalies | Photoplethysmography (Blood Flow) | Grayscale Pixel Manipulation |
| Claimed Accuracy | 91% | 96% (Real-time) | Variable |
| Key Advantage | Community-driven updates; localized for diverse facial structures | Extremely fast real-time analysis | Strong integration with enterprise cloud systems |
| Target Audience | Global Developers & Researchers | Broadcasters & Large Tech Platforms | Enterprise Security & Government |
The Open-Source Debate: A Double-Edged Sword?
By choosing to make SARAAB open-source, Dubai’s DESC has sparked an important conversation within the global security community.
The Advantages of Open-Sourcing
- Global Collaboration: Hosting SARAAB on Hugging Face allows thousands of independent AI researchers to test, debug, and improve the model’s codebase. This collective intelligence can help update the model much faster than a closed team could.
- Transparency: Proprietary systems keep their detection methods secret, making it hard for independent journalists or researchers to verify their claims. SARAAB offers a transparent, auditable tool for public trust.
- Reduced Costs for Enterprises: Small-to-medium businesses (SMBs) that cannot afford expensive enterprise security suites can integrate SARAAB’s code directly into their security pipelines via APIs.
The Risks of Open-Sourcing
- Reverse-Engineering: The primary risk of open-source security tools is that malicious actors can download the exact code. By understanding exactly how SARAAB detects deepfakes, bad actors can train their generative adversarial networks (GANs) to bypass SARAAB’s detection thresholds.
- The Cat-and-Mouse Game: As soon as SARAAB’s detection code is published, deepfake creators will find ways to work around it, requiring constant, rapid updates to the model to keep it effective.
Pros & Cons of the SARAAB AI Model
Pros
- Granular Detection: Capable of flagging highly localized, short-duration manipulations within long videos.
- Visual Proof: The heat map feature provides clear, visual evidence of tampering, which is highly useful for legal and journalistic verification.
- Sovereign Innovation: Developed entirely by Emirati talent, reducing reliance on third-party foreign security software.
- Free Accessibility: Open-source licensing lowers the barrier to entry for academic researchers and budget-conscious organizations.
Cons
- Unverified Benchmarks: The 91% accuracy rate is currently self-reported by DESC; independent validation on public datasets is still pending.
- Bypass Vulnerability: Open access allows bad actors to study the model’s limitations and engineer smarter, undetectable deepfakes.
- Evolving Threat Landscape: No single model can act as a permanent solution to deepfakes, as synthetic generation technologies continue to evolve rapidly.
What SARAAB Means for Enterprise Security
For businesses operating in 2026, deepfakes are no longer just a political or social issue. They represent a direct financial threat.
Cybercriminals increasingly use high-quality synthetic audio and video to conduct business email compromise (BEC) scams. They impersonate CEOs in virtual meetings and execute sophisticated corporate fraud.
Once SARAAB is launched on Hugging Face in late 2026, corporate IT departments can use its API to build internal verification systems. For example, video files uploaded for identity verification (KYC) in banking, or recorded video instructions for high-value wire transfers, can be automatically routed through SARAAB to verify authenticity before processing.
The Verdict
Dubai’s announcement of SARAAB at GISEC Global 2026 marks a bold step in the global fight against synthetic misinformation. By offering a highly capable, 91% accurate detection tool to the open-source community, DESC is encouraging global collaboration to solve a shared digital challenge.
While the risk of reverse-engineering by bad actors remains real, the benefits of transparent, community-driven security tools generally outweigh the drawbacks. As SARAAB nears its public release on Hugging Face in late 2026, it will serve as an important benchmark for how governments can actively contribute to global digital trust.
