The EU's New Brussels Team Will Police AI Deepfakes and Label Chatbot Content

Key points

• The EU's AI Act will require companies to label or watermark AI-generated content, including chatbot replies and images.
• The European Union is establishing a dedicated enforcement unit in Brussels.
• The rules cover deepfakes, realistic AI-fabricated videos or images of real people, and AI-generated imagery used for illegal purposes.
• Businesses deploying AI tools for consumers must meet the labelling requirements before the relevant provisions take effect.
• The Brussels unit's remit also extends to AI-assisted hacking.

Europe is setting up a specialist enforcement unit in Brussels to police artificial intelligence, and one of its first jobs will be making sure consumers know when they're looking at AI-made content. Under the EU's AI Act, companies will be required to attach labels or digital watermarks to chatbot responses and AI-generated images so users can tell what a machine produced.

Passed into EU law in 2024, the Act's provisions are taking effect in stages. Labelling requirements sit within the transparency obligations section and are among the rules consumers are most likely to notice.

Why does this matter to ordinary people?

A deepfake is a video or photo of a real person fabricated by AI, often to make them appear to say or do something they never did. Scammers and propagandists use them. Required labelling means a reader scrolling social media or using a customer-service chatbot should see a clear signal that the content came from software, not a person. Think of it as a nutritional label for digital content.

For that to work, the enforcement unit needs real authority to pursue companies that skip the label. That's the gap the Brussels team is meant to close.

Our 30 July story on deepfakes built from school photo feeds showed how quickly the harm scales when fabricated imagery goes unlabelled. A disclosure requirement with actual enforcement behind it is a meaningful step, though the Act's staged rollout means some obligations are still months away from biting.

Should you worry?

If you see AI-labelled content online, treat it the way you'd treat an advertisement. It was built by software and that's worth bearing in mind before sharing it or acting on it.

What should businesses do now?

Companies offering AI chatbots or producing AI-generated images for consumers in EU member states should be reviewing their products against the Act's transparency requirements now, not after an enforcement notice arrives. The obligations fall on providers, meaning the companies that build and deploy these tools, not only the platforms that host them.

Any business collecting personal data through an AI system also needs to consider how labelling intersects with the GDPR, the EU's General Data Protection Regulation, which gives individuals rights over how their personal information is used.

Reported first by SecurityWeek, the Brussels unit's remit also covers AI-assisted hacking, which puts the team's scope well beyond consumer labelling alone.