EU rules now require chatbots to say they are AI
On Sunday, 2 August 2026, the transparency part of the EU AI Act started to apply. The European Commission marked the date with a short notice: from now on, people in the EU must be told when they are dealing with an AI system instead of a person, and some content made by AI has to carry a label.
This is the part of the law most people will actually run into. It covers the chat window on a store's website, the voice agent on a support line and the avatar inside an app.
What the rule asks of a chatbot
The Commission names chatbots, AI agents and avatars as examples of systems that must disclose they are AI. Its FAQ on Article 50 (the article of the AI Act that holds these duties) says the notice has to come at the start of the first interaction, in a clear form that also meets accessibility requirements. When it is obvious to an average, reasonably well-informed person that they are talking to a machine, no notice is needed. The Commission says that exception should be read narrowly.
Labels are the second half. Deepfakes, meaning images, audio or video that look like real people, places or events, must be labeled. So must text published to inform the public on matters of public interest when no person reviewed or edited it. The EU has drawn a set of icons for these labels, and providers of generative AI also have to add machine-readable marks so software can detect the content.
What can wait until December
Systems that were already on the market before 2 August have until 2 December 2026 to add the machine-readable marking. Content made before 2 August does not need a label after the fact. The broader rules for high-risk uses, such as AI in hiring, education or critical infrastructure, now apply from 2 December 2027, after the changes made to the law earlier this year.
Who checks, and how big the fines are
National market surveillance authorities enforce the rules, and the EU's AI Office covers the systems under its own supervision. Fines for companies go up to €15 million or 3% of global annual turnover. The Commission says it applies proportionality to small and medium-sized businesses, and it does not give a separate figure for them.
What to do this week
If you run a chatbot or a voice agent that serves people in the EU, open it as a new user and read the first message. If that message does not say it is an AI, that is the line to change. The Commission's guidelines on Article 50 and the voluntary code of practice on labeling are the two documents to read next.
What we think
We like this rule. Telling people up front that they are talking to an AI costs one sentence, and it spares the person on the other side from guessing. The labeling of public-interest text is where we expect the arguments, because the line between reviewed by a person and not reviewed will be drawn case by case.
Sources