AI Act fines: what you actually risk

Up to €15m or 3% of worldwide turnover. The real issue is inspection, formal notice, and proof.

AI Act sanction ceilings are high. They mainly target serious breaches. The immediate risk is operational: having to explain, quickly, what the chatbot does and how users are informed.

Keeping a dated log of displayed notices, a register of AI tools and scan dates is more useful than a one-off report forgotten in a folder.

General information, not legal advice.

More guides

AI Act Article 50: what you must disclose
Since 2 August 2026, chatbots and AI-generated content must be identifiable. Timeline, duties, and proof.
AI chatbot: the disclosure duty since 2 August 2026
A support widget is not “obvious”. Here is the notice to show, when to show it, and how to prove it.
Marking AI-generated content (C2PA, metadata)
Article 50(2) requires machine-readable marking. Transition until 2 December 2026 for systems already online.
2 December 2026: end of the AI marking transition
Last window before mandatory machine-readable marking of existing systems, then high-risk systems in 2027.
Proving AI transparency: what an inspector may ask for
A visible notice is not enough. The useful file is a register, a display log, and dated scans — not a legal opinion.
AI system register: what to record for your site
Chatbot, copy, images: a simple inventory (system, vendor, use) avoids a blank stare when someone asks what you run.
Crisp and AI Act Article 50: who must show the notice?
Crisp provides the widget. The site must tell visitors they are talking to AI, at the first interaction.

AI Act fines: what you actually risk

More guides

AI Act Article 50: what you must disclose

AI chatbot: the disclosure duty since 2 August 2026

Marking AI-generated content (C2PA, metadata)

2 December 2026: end of the AI marking transition

Proving AI transparency: what an inspector may ask for

AI system register: what to record for your site

Crisp and AI Act Article 50: who must show the notice?