The EU now requires disclosure when AI content looks real

Signals Inbox·August 3, 2026·AI Trust

Europe has moved AI disclosure from a voluntary trust feature to a legal requirement. Since 2 August 2026, certain chatbots, deepfakes, realistic synthetic media, and AI-written public-interest content must tell people what they are dealing with. The real shift is that AI provenance is now becoming part of the product itself.

Q1What actually changed?

The EU AI Act’s Article 50 transparency rules started applying on 2 August 2026. People must now be told when they are speaking with certain AI systems or seeing specific types of AI-generated or manipulated content.

Q2What content needs a label?

The rules cover several different cases. Chatbots must normally identify themselves as AI. Deepfakes and realistic synthetic images, audio, or video must be disclosed. AI-written text about matters of public interest also needs disclosure when it is published without human review or editorial control.

Q3Does every AI image need a warning?

No. The rule is not simply that every edited photo needs a giant AI sticker. The strongest disclosure duty targets content that could mislead people by looking like real people, places, objects, or events. Artistic and fictional work can receive a lighter disclosure that does not ruin the experience.

Q4Why does this matter now?

Because realistic AI media has become cheap, fast, and difficult to spot with human eyes alone. Europe is responding by making disclosure part of distribution rather than expecting every viewer to become a deepfake detective. That turns provenance, watermarking, and content credentials into real product requirements.

Q5Who has to do the work?

Both sides of the stack. AI providers must make synthetic outputs detectable in a machine-readable way when required. Companies and professionals using those systems must add clear disclosure when they publish covered deepfakes or public-interest content. A hidden mark inside the file may not be enough for the person viewing it.

Q6So what changes for AI companies?

Disclosure can no longer be treated like a line buried in the terms of service. Chatbot notices, visible labels, metadata, content credentials, audit logs, and human-review workflows now need to be designed into products. Some older systems placed on the market before 2 August receive a transition until 2 December 2026, but the direction is already clear: believable AI needs receipts.