EU AI Act — General Purpose AI Model obligations now in force for providers above threshold
What the law is now
The EU AI Act imposes tiered obligations on providers of AI systems classified by risk. The General Purpose AI chapter covers providers of models that power multiple downstream uses. As of August 2025, all GPAI model providers serving the EU must meet baseline transparency obligations. Providers whose models reach the systemic risk threshold (currently 10^25 FLOP training compute) face additional adversarial testing and incident-reporting requirements.
What just shifted
What this adds: The August 2025 effective date moved GPAI obligations from a future compliance target to an active legal duty. Providers must now maintain technical documentation, publish a summary of training data including copyright policy and opt-out compliance, and put in place downstream deployer information obligations. Providers above the systemic risk threshold must additionally conduct model evaluations, red-team testing, and notify the AI Office of serious incidents.
What this puts in question: Whether companies that fine-tune or adapt a GPAI model for their own product are themselves GPAI providers subject to these obligations, or whether only the original model developer bears them — the AI Office's codes of practice are expected to clarify this boundary.
What clients should weigh
Watch for
· European AI Office GPAI code of practice, final version expected Q4 2026
· AI Office designation of which existing models meet the systemic risk threshold
· UK AI Regulation Bill and its relationship to the EU AI Act for UK-EU dual-operating companies
Ready to use
These are drafts. Edit before sending to a client.
Client alert
Draft — edit before sending to a client.
This corpus reflects one attorney's personal review. It is not a comprehensive survey. Verify scope and currency before relying on it for any matter.