EU Enforces AI Act with Fines Reaching Up to 7% of Sales

On August 2, 2026, the European Union announced the activation of the final phase of the AI Act, which covers General Purpose AI (GPAI) models used in various applications such as digital assistants and content generation systems.

The law applies retroactively to August 2025, meaning that companies that violated its provisions during that period could face immediate financial penalties. Fines range from 15 million euros or 3% of global sales for general violations, rising to 35 million euros or 7% for strictly prohibited practices such as psychological manipulation of users.

The law imposes specific obligations on GPAI model providers, including disclosing the AI's identity to users, embedding machine-readable digital watermarks in generated content, and submitting detailed transparency files to regulators.

This enforcement represents a global regulatory precedent, as major tech companies like Google, Meta, and OpenAI await the first enforcement action by the European AI Office — which now holds the authority to directly investigate and impose fines without cases having to go through national courts first.

What do these terms mean?

GPAI Models: General-purpose AI models like ChatGPT and Gemini are trained on massive datasets and perform multiple tasks, unlike models specialized in a single task.

Prohibited Practices: Include psychological manipulation of users, biometric discrimination, and social scoring of individuals by government authorities — and are subject to the highest fine categories under the law.

Conformity Assessment: A mandatory procedure in which a company proves that its product meets the law's standards before launching it in the European market, similar to safety certifications required for electrical appliances.