EU AI Act enforcement begins: Brussels gains power to fine and pull GPAI models

On August 2, 2026, the European Commission’s AI Office gains power to investigate, fine and order the market withdrawal of general-purpose AI (GPAI) models under the EU AI Act.

The transparency, copyright and systemic-risk obligations on GPAI providers have applied since August 2, 2025, but Brussels could not act on breaches — it could not demand documents, order fixes or levy fines until the one-year grace period closed. That gap closes on August 2. The Commission said national authorities and the AI Office begin full supervision and penalty enforcement the same day.

Under Article 91, the AI Office can compel technical documentation from providers; under Article 92 it can run its own technical evaluations, require independent expert access and order compliance fixes or risk mitigation. In the most serious cases it can restrict, withdraw or recall a model from the EU market. Fines for GPAI non-compliance reach 3% of global annual turnover or 15 million euros, whichever is higher; penalties for prohibited AI practices climb to 7% or 35 million euros.

A second set of rules takes effect the same day. Chatbots and interactive systems must tell users they are AI, deepfakes must be labeled, and AI-generated or altered content must carry machine-readable marks. More than 180 organizations have signed a Code of Practice on AI-content transparency, the Commission said.

The powers are broad but untested. No provider has yet been investigated under them, and the Commission has not said which models or companies it will scrutinize first. Whether Brussels moves quickly against a frontier model or opens with documentation requests will signal how hard the AI Office intends to bite.