AI Act fines: how high they are and what they are really for

A lot of numbers circulate about AI Act fines, and most are quoted without context. For the obligations that affect an ordinary company, the regulation has three different levels.

Prohibited practices: €35 million or 7% of turnover

The highest rate in the regulation applies to Article 5, the short list of things that are banned outright. It has applied since February 2025.

For an ordinary company, one item on it really matters: emotion recognition in the workplace. A camera that assesses whether people look focused. Software that reads a candidate’s facial expressions in an interview and judges whether they seem trustworthy. The workplace is interpreted broadly and includes remote work and hiring. The only exceptions are medical and safety-related.

Companies usually end up here by accident. Someone buys a productivity monitoring tool without knowing exactly what it does.

Transparency: €15 million or 3% of turnover

Article 50 has applied since 2 August 2026. A chatbot that talks to customers has to make clear right at the start of the conversation that it is a machine. Generated images, audio or video that look like a real recording must be labelled when published.

Breaches are capped at €15 million or 3% of worldwide turnover. For SMEs, the lower of the two amounts applies.

Untrained staff: no separate fine in the regulation

The regulation sets no separate fine for failing Article 4 on AI literacy. Penalties are set by national authorities.

What costs more than a fine

The fine is the less likely part of the problem. It tends to cost more when a client contract or personal data ends up in a public chatbot, which is also a data protection issue with its own penalties. When an offer or report goes out with a figure the AI made up. When an employee sends money on an instruction from a cloned voice of the managing director. And when a client asks how you have AI covered and you have nothing to show.