The European Commission's enforcement and penalty powers over general-purpose AI (GPAI) model providers become applicable on August 2, 2026, converting a year of quiet obligation into a live regulatory regime with real teeth. From that date, the EU AI Office can compel documentation under Article 91, run technical evaluations under Article 92, demand risk-mitigation measures under Article 93, and issue fines of up to 3% of global annual turnover or €15 million, whichever is higher, under Article 101 of the AI Act.
What Actually Changes On August 2
GPAI obligations themselves are not new — they applied from August 2, 2025, alongside the GPAI Code of Practice. What the Commission held back for twelve months was the enforcement machinery. That grace period ends today. The Commission can now also restrict or withdraw a model from the EU market, and refusing or stalling on a documentation or evaluation request is itself a finable offense.
For clarity: the 3% / €15M ceiling covers GPAI infringements. The much-quoted 7% / €35M maximum applies to prohibited AI practices under Article 5, not to GPAI documentation or transparency failures.
The Code Of Practice Divide
The voluntary GPAI Code of Practice is now the fastest route to a presumption of conformity with matching Transparency, Copyright, and (for systemic-risk models) Safety and Security obligations. The industry has split visibly:
Signed the full Code: Anthropic, Google, Microsoft, OpenAI, IBM, Mistral AI, Cohere and Amazon. xAI signed only the Safety and Security chapter. Meta declined to sign, citing legal uncertainty. Meta and xAI are not breaking the law — the Code is voluntary — but they have taken on the burden of demonstrating conformity by “other adequate means” and answering the AI Office directly, rather than relying on the Commission's presumption.
Digital Omnibus Does Not Rescue GPAI
In May 2026 EU negotiators reached a provisional agreement on the “Digital Omnibus,” pushing several high-risk deadlines back — stand-alone high-risk Annex III systems now slip to December 2, 2027, and high-risk AI embedded in Annex I regulated products moves to August 2, 2028. That relief does not touch GPAI. The August 2, 2026 enforcement date for model providers stands, and pre-2025 GPAI models must still be brought fully into compliance by August 2, 2027.
Why It Matters For AI And Robotics Companies
For anyone building physical AI, autonomous systems, or agentic products on top of frontier models in the EU, vendor compliance posture is now part of your compliance posture. If the AI Office restricts a model in the EU, downstream products that depend on it are exposed. Enterprises are being urged to inventory model dependencies, ask vendors for their Code status and conformity approach, and design for model portability. Related coverage: IBM Claims Verified Quantum Advantage, Apptronik Apollo 2 and Robot Park, and AMD and Anthropic's $5B Equity Deal.
Reporting based on coverage from Beam AI, the European Commission, DLA Piper, Gibson Dunn and Latham & Watkins analyses of the AI Act.