AI labeling rules start August 2: what actually changes for you

New EU transparency rules for artificial intelligence take effect in one week, on August 2. They sit in Article 50 of the AI Act, and they affect you directly. Chatbots have to identify themselves as machines, and deepfakes have to be disclosed. The legal groundwork only just fell into place. Regulation (EU) 2026/1744 appeared in the Official Journal on July 24 and enters into force on July 27, and the European Commission published its guidelines on how to read Article 50 on July 20, less than two weeks before the deadline.

The four duties in Article 50

Providers have to build chatbots and voice assistants so that people can tell they are talking to a machine. They also have to mark the output of generative AI in a machine-readable format, which means adding a technical marker to images, audio, video and text. Anyone who publishes a deepfake has to say so, and that covers deceptively realistic content featuring real people. The same applies to AI-generated text published to inform the public on matters of public interest, unless a person has reviewed it and takes editorial responsibility.

There are exceptions. If the AI only assists with standard editing, such as fixing spelling and grammar, the marking duty does not apply. Content that is obviously fantastical falls outside the definition of a deepfake as well. The Commission gives people flying through the air as its example.

Why you will barely see any labels

Machine-readable is not the same as visible. Article 50(2) is about watermarks and metadata that detection tools can read, not about an icon that catches your eye. There is also a transition period. Generative systems that were already on the market before August 2 get four extra months for that marking, until December 2, 2026. In practice that covers just about every service you use today.

A visible label does exist. The Commission offers three icons, one for content that AI had a hand in, one for fully generated content and one for content that was only partially modified. Using them is voluntary and part of the Code of Practice on transparency of AI-generated content. The icon on its own proves nothing. The duties under Article 50 apply either way.

Who enforces this and what breaches cost

In Germany the Federal Network Agency becomes the central market surveillance authority, and the implementing law cleared the Bundesrat on July 10. The agency is setting up a contact point where you can report suspected breaches. For breaches of the transparency duties, the AI Act allows fines of up to 15 million euros or 3 percent of worldwide annual turnover. How firmly the EU pushes its rules on the big providers is on show in the ongoing dispute over AI assistants on smartphones.

Two new bans arrive later. From December 2, 2026, AI systems that produce non-consensual intimate imagery or child sexual abuse material may no longer be offered in the EU. Those sit in the top penalty tier of up to 35 million euros or 7 percent of annual turnover. The stricter rules for high-risk AI, by contrast, have been pushed back to December 2027 and August 2028.