EU AI Act's general-purpose AI obligations enter into force
On August 2, 2025, the European Union's AI Act began applying its core obligations to providers of general-purpose AI (GPAI) models, and the EU's AI Office took up its role as the coordinator of enforcement, according to the European Commission and reporting by Reuters and POLITICO. The new duties cover transparency, copyright compliance, and technical documentation, and apply additional safety obligations to models trained above a 10^25 floating-point operations (FLOPs) compute threshold deemed to pose "systemic risk."
The AI Act, formally adopted in 2024, layers obligations onto GPAI providers in stages. Models placed on the EU market after August 2 must comply immediately; models already on the market before that date have until August 2, 2027, to come into line, according to guidance published by the European Commission and DLA Piper's August 2025 client alert. The Commission's full enforcement powers, including the ability to impose fines on GPAI providers, begin to apply on August 2, 2026.
In the run-up to the deadline, the Commission published a voluntary Code of Practice meant to ease compliance. Amazon, Anthropic, Google, IBM, Microsoft, OpenAI, Mistral AI, and Aleph Alpha signed; xAI signed only one of the three chapters; Meta declined. In a July 18 LinkedIn post, Meta global affairs chief Joel Kaplan wrote that "Europe is heading down the wrong path on AI" and that the Code "introduces a number of legal uncertainties for model developers, as well as measures which go far beyond the scope of the AI Act," in remarks reported by The Wall Street Journal and POLITICO.
The penalty regime that began applying on August 2 sets administrative fines of up to EUR 35 million or 7% of worldwide annual turnover for prohibited AI practices, up to EUR 15 million or 3% for other obligations, and up to EUR 7.5 million or 1% for supplying incorrect or misleading information, per DLA Piper's published summary. Penalties specific to GPAI providers, however, are not enforceable until August 2, 2026. The Commission set the systemic-risk compute threshold at 10^25 FLOPs.
Industry reaction ranged from cautious endorsement to pushback. The Computer & Communications Industry Association, of which Meta is a member, lobbied for delays, but Commission spokesperson Thomas Regnier told reporters on July 4, "Let me be as clear as possible, there is no stop the clock. There is no grace period. There is no pause," according to Reuters. Lawyers at Arnold & Porter and DLA Piper described the rollout as the most consequential AI regulatory milestone of 2025.
For us at Enpo Sekai, the GPAI obligations matter less for our day-to-day output — we do not train foundation models — and more for the supply chain we depend on. Several of our character and persona pipelines run on third-party GPAI providers, and the new transparency, copyright, and documentation duties placed on them will shape what data and model-provenance information we can request, audit, and pass through to our own customers and players. We will adjust our vendor agreements accordingly.
