Articles

Six pieces on what Article 50 and the Commission's guidelines actually say. Everything is quoted directly and linked to the primary source, so you can check us rather than trust us.

The EU didn't delay the rule that applies to your chatbot

The Digital Omnibus moved the high-risk obligations to 2027 and 2028. It amended Article 50 exactly once, at paragraph 7, which is about codes of practice. Here is how to verify that in three documents and ten minutes.

What counts as an AI disclosure, according to the Commission

Paragraphs (37) and (38) of the final guidelines: one list of techniques that work, one list of techniques that do not. Terms and conditions are on the second list. So is calling it an "assistant".

"It's obviously a bot"

There is a real exception where AI use is obvious. The Commission reads it restrictively — "almost no doubt left" — and names helpdesk chatbots on online platforms as an example that does not qualify.

Provider or deployer? The question agencies keep getting wrong

Article 50(1) binds providers; (3) and (4) bind deployers. And the definition of provider covers systems supplied "whether for payment or free of charge", which closes the escape route most agencies reach for first.

Ask your chatbot if it's human

The guidelines require disclosure whenever a user asks what they are talking to. Almost nothing ships with this, and no snippet can fix it — it lives in the system prompt. Includes the prompt, in English and Polish.

€15M or 3%? Read Article 99(6) first

The headline fine is real. For SMEs, paragraph 6 turns "whichever is higher" into "whichever is lower", and nobody quoting the headline mentions it.

Who actually enforces this?

One regulation, 27 member states, and the words "at least one" in Article 70. Some states designated a single AI authority; others spread it across their sectoral regulators. Includes what we are deliberately not publishing yet, and why.