France Riots: Police Facial Recognition on Social Media [2026]

French investigators are turning to facial recognition on social media images to identify people who took part in last weekend's riots. According to Biometric Update's October 22, 2026 report, Justice Minister Gérald Darmanin announced in a televised interview that police will compare faces against the millions of face profiles in the TAJ criminal records file and against images pulled from platforms including TikTok, Instagram and Discord. Roughly 200 investigators are assigned nationwide, and a dozen specialist officers are running the face searches, the report says, citing France24.

The announcement lands in a country where police face recognition has been a legal minefield for years. The data protection authority, the CNIL, has already found problems with how the gendarmerie used BriefCam software. This post explains what is known, what the law allows, why the social media angle is the part that worries privacy experts, and what to do if you film or post crowds.

What Exactly Did French Police Announce?

French police will use facial recognition to match riot suspects against the TAJ and social media images, and are reviewing footage from about 80,000 security cameras. Biometric Update's report says investigators are examining video from cameras across Paris streets and public transit as well as shop cameras, and are also acting on incidents not reported to them by using social media and geolocation.

One officer told France24: "it's not the only tool to close a case, but it's one of the very useful tools for investigators." The framing matters. French authorities present facial recognition as one aid among many, used on footage after the fact rather than as a live scanner.

The report does not give arrest numbers, the cities affected or the cause of the unrest, and we have not independently verified those details. Treat the facts above as what officials said, as relayed by Biometric Update and France24.

What is the TAJ?

The TAJ is the national file of judicial records used by the police and gendarmerie. It contains photographs of people who have been implicated in investigations, and it is the main reference set for face matching. Because the file includes people who were never convicted, the size and accuracy of the face matching pool is a recurring concern for digital rights groups.

Is Police Facial Recognition Legal in France?

After-the-fact comparison of faces inside a judicial investigation is permitted in France, but real-time use and unregulated tools have been ruled unlawful. Biometric Update says police use of facial recognition for forensic identification is enshrined in French law and has survived challenges at the EU level.

The controversy is about where that stops. In 2023, the outlet Disclose reported that national police had used BriefCam video analytics software with face recognition features since 2015. The CNIL and the Interior Ministry investigated. According to Next's coverage, the CNIL's decision of November 15, 2024, published in the Journal officiel on December 5, 2024, put the Interior Ministry on formal notice to take adequate measures to block facial recognition functions in the judicial matching software it uses. Biometric Update adds that the CNIL found the gendarmerie deployed the technology in public without approval or completed data protection impact assessments, which made those deployments unlawful.

In 2025, La Quadrature du Net reported that a Grenoble administrative court found BriefCam processing in the town of Moirans disproportionate and not provided for by law. That organization is an advocacy group, so weigh its framing accordingly, but the ruling itself is relevant to how far retrospective video analytics can go.

Earlier in 2026, Disclose reported that French police used Cognitec face biometrics for real-time checks against the TAJ, allegedly contrary to the Code of Criminal Procedure. The gendarmerie denied the allegations, according to Biometric Update. In other words, the legal basis for this week's announcement exists for retrospective forensic comparison, while the boundaries around tools and real-time use remain contested.

What does EU law say?

The EU AI Act restricts several uses of biometric identification and bans building facial recognition databases through untargeted scraping of images from the internet or CCTV. It treats law enforcement use differently, with narrow exceptions and safeguards for targeted searches in serious cases. For the compliance detail, see our guides to the EU AI Act scraping ban and retrospective facial recognition on CCTV archives.

Why Do Social Media Images Change the Picture?

Social media turns every public post into a potential face search input. CCTV and police files are limited sets. Platforms hold billions of images that are already labeled with names, places and friends. When investigators combine a face matching engine with a platform image, the pool of people who can be identified expands from people already in police files to anyone who has been photographed and posted.

Two details make this worth attention:

• Bystanders are included. A search for a suspect in a crowd can return other faces. People who filmed, watched or happened to be on a street could be matched, correctly or not.
• Error has consequences. Face matching produces candidate lists, not proof. Our reporting on the WA Police live facial recognition trial shows how matches depend on thresholds and human review.

The same dynamic is visible beyond France. Germany's lawmakers proposed changes to the code of criminal procedure earlier in 2026 that would let police run social media images through facial recognition; they have not been approved. The UK courts are separately reviewing Clearview AI's scraped face database.

Does Blurring Faces Help If You Film a Riot or Protest?

Yes, for the footage you publish. Blurring faces in a clip you upload means that clip cannot be used to match the people you blurred. It does not remove them from other footage, and it does not protect you from CCTV. If you film demonstrations, riots or anything involving crowds, think of blurring as controlling your own contribution to the pile.

BGBlur is a browser-based tool that detects faces and keeps the blur locked on as people move, so you do not need to edit frame by frame. It can also blur license plates, which matters if your clip shows vehicles, and anonymize voices if people are speaking. You can try the free tier (up to 200MB and 10 minutes per video, three videos a month, 720p export), and Pro removes the watermark and limits. Videos are deleted within 24 hours.

The caveats are real. Blurring does not hide clothing, tattoos, gait, voice or location. A blurred face with an unusual jacket in a recognizable street is still identifiable to people who know it. If you post original unblurred footage elsewhere, the protection is gone. And blurring cannot undo what has already been posted: weak pixelation can sometimes be reversed, as we explain in our guide to why weak blur fails.

For protest-specific advice, see how to protect privacy at protests from drones and body cameras.

What Should You Do If You Filmed Last Weekend's Events or Similar Footage?

If you have crowd footage, the safest approach is to decide what you need it for before you post it. Here is a numbered plan:

1. Decide the purpose. Documenting police conduct or public safety is different from sharing dramatic clips for views.
2. Do not tag or name people. Naming turns a face into an identification.
3. Blur bystanders and minors. Use motion-tracked face blur so every frame is covered.
4. Check the blur. Scrub through the output at normal speed and frame by frame for slips when someone turns or leaves the frame.
5. Remove metadata. Strip location data from stills before you post.
6. Keep the original private. Store it offline in case you need to give it to a lawyer, a journalist or an authority.
7. Know your rights. In France and the EU, you can ask a controller about data it holds on you under GDPR and the French Data Protection Act, with exceptions for criminal investigations.
8. Take down old posts if needed. If something you posted is misused, report it to the platform and consider contacting the CNIL.

For the wider GDPR framework, see our GDPR video compliance guide.

How Accurate Is Facial Recognition on Riot Footage?

Accuracy drops sharply on the kind of images riots produce. Face matching works best on well-lit, front-facing photos, such as the custody images in a police file. Riot footage is the opposite: masks, hoods, smoke, motion blur, low-resolution CCTV and phone video shot from the side or above. A search engine will still return a ranked list of candidates, but the quality of the underlying image determines how meaningful that list is.

That is why the human step matters. In the French account, specialist officers run the searches and investigators then corroborate with other evidence such as location data, clothing and social posts. The officer's remark that facial recognition is "one of the very useful tools" but "not the only tool to close a case" reflects that division of labor. The risk is not the tool alone, but a candidate list that investigators treat as more certain than it is, particularly under political pressure to deliver arrests quickly after a high-profile weekend.

For people on the receiving end, the practical issues are these:

• A wrong match can still trigger questioning. Even if it never reaches court, being named as a candidate can mean a visit, a search or a social media pile-on.
• Aged or low-quality images raise false match risk. An older photo in a file and a grainy frame from a transit camera make a poor pair.
• Group posts amplify errors. If a mistaken name circulates online, a later correction rarely travels as far.

None of this is unique to France. The same limits apply to any face search, from police systems to commercial services like the ones covered in our post on Clearview AI and the UK ICO.

What Rights Do People Have Over Their Faces Online?

In the EU, a face template is biometric data and receives special protection under the GDPR, but law enforcement processing falls under a separate directive with broader exceptions. Ordinary photos are personal data. Biometric data used to uniquely identify a person is a special category, and the default is that it cannot be processed unless a narrow condition applies. Police have their own regime, which allows processing that is strictly necessary and subject to safeguards, with oversight from the CNIL.

For ordinary users, three rights are most useful:

1. Right of access. You can ask who holds data about you and what it is. Exceptions apply for ongoing investigations.
2. Right to erasure. You can ask a platform to remove images you posted, and in some cases images others posted of you, particularly where consent was missing.
3. Right to complain. The CNIL accepts complaints about unlawful processing, including by public authorities.

Rights are slower than blur. If you want an image never to be searchable, the strongest control is not publishing a recognizable version. Our overview of UK GDPR and the Data Protection Act for video privacy explains how similar rules apply across the Channel.

What about journalists and documentary filmmakers?

Reporters and documentary makers face a real tension. Showing faces can be central to public interest reporting, but publishing identifiable crowds can expose sources and bystanders. A defensible approach is to show faces only when the person is the subject of the story or consented, and to blur the rest. Newsrooms increasingly keep an unblurred master for legal and editorial purposes while publishing a blurred version. BGBlur fits that workflow: it exports a blurred copy and leaves your original untouched, and videos processed on the service are deleted within 24 hours.

What Should Observers Watch Next?

Watch for four developments. First, any official figures on arrests and on how many identifications came from face matching versus traditional methods. Second, whether the CNIL or courts examine the new searches, given the November 2024 formal notice. Third, whether platforms respond to requests for images. Fourth, whether the French government moves to change the legal framework for police facial recognition.

Conclusion

France's riot investigations are a case study in the gap between what is legal and what is comfortable. Police face matching on archives and social media images is allowed in the forensic setting, yet the regulator has already found that earlier deployments broke the rules, and courts have questioned related tools. Because the pool for matching now includes platform photos, ordinary posts become part of the evidence pipeline.

You cannot control CCTV, but you can control what you publish. Blur the faces of people who did not choose to be in your video, avoid tagging, and keep originals private. BGBlur can do the blurring in your browser with motion tracking, without a download, so you can document an event without turning bystanders into search results. As governments across Europe debate similar powers, the sensible default is to treat every public image as searchable and to publish with that in mind.