Arrested French Protesters on Video? Fact-Checkers Say It's AI
A video appeared to show people arrested in French protests being sentenced in court. It spread across Facebook, Instagram, YouTube and X. Fact-checkers at Myth Detector say it was AI-generated, and their method is a useful model for anyone who has to verify a viral clip.
Fact-checkers do not rely on a single test. They inspect the footage, trace its source and use AI detection tools as supporting evidence.
What Myth Detector found
On 26 October, a Facebook user posted a French-language video made of several scenes that appeared to show verdicts being read out to people detained in the protests in France. Myth Detector, a Georgian fact-checking organisation, concluded that the footage was created with AI. It based that on visual errors in the video and on the results of AI detection tools.
The clip circulated widely on 24 and 25 October across several platforms. Myth Detector traced earlier copies to a TikTok account that had posted the footage on 17 October, which looks like the likely original source. That account had published dozens of other videos with scenes of the same kind.
The video appeared during weeks of protests in France, a period in which, according to Myth Detector, fake and manipulated images and videos have often circulated.
The visual clues fact-checkers pointed to
Skin that looks too smooth
In the first scene, the detained person's skin lacked pores and natural texture.
Camera details that don't add up
The angle of a police officer filming the accused, and how the accused appears on the phone screen, did not match reality. In another scene, the image on a spectator's phone did not fit what zooming in showed.
Hands and tears
A man's tears and his hands over his face looked off, with hands out of proportion to his head. Myth Detector notes AI models often make this kind of error.
Speech that doesn't match the mouth
A voice said he had done it to feed his children, yet the accused's mouth did not move while the words played.
Where AI detection tools fit
Myth Detector also ran the footage through AI detection tools, UncovAI and Hive Moderation, and reports that the results showed traces of AI use in the visuals.
That sequence matters. The fact-check does not start from a score. It starts from what is visible and where the video came from, and the tools add independent evidence. For newsrooms that need that second opinion on a clip, UncovAI's video detection is built for it.
A detection result is one input. The conclusion should rest on several kinds of evidence that point the same way.
A checklist for verifying a viral video
Trace the earliest copy. Look for the first account that posted it and what else that account publishes. A feed of near-identical scenes is a warning sign.
Watch it frame by frame. Check skin, hands, text, reflections and anything shown on a screen inside the video.
Match sound to picture. Does the mouth move with the words? Does the audio fit the setting?
Cross-check the claim. Look for reports from reliable outlets, official records or other footage from the same event.
Run a detector. Use the best-quality copy you can find, and read the result alongside everything above. If you work in a newsroom, see who UncovAI is built for.
Questions about AI-generated video
How do fact-checkers verify an AI-generated video?
They combine three things: a close visual inspection of the footage, tracing the video back to its earliest source, and AI detection tools. No single step is treated as enough on its own.
Can an AI detection tool confirm a video is fake by itself?
No. A detector gives an estimate that supports an investigation. Fact-checkers pair it with visual evidence and source tracing before reaching a conclusion.
What visual signs suggest a video is AI-generated?
Examples from this case include unnaturally smooth skin, body parts that look out of proportion, details on a phone screen that do not match the scene around it, and speech that continues while the speaker's mouth stays still.
What is Myth Detector?
Myth Detector is a Georgian fact-checking organisation. Its website lists it as a signatory of the International Fact-Checking Network (IFCN), and this article was prepared within Facebook's fact-checking programme.
Verify before you publish
Test a video, image or audio clip with UncovAI before it goes out.
Get Started Free →