EU Issues First-Ever AI Safety Guidelines for High-Risk Models

EU Issues First-Ever AI Safety Guidelines for High-Risk Models

The European Union published a General-Purpose AI Code of Practice on July 10, 2025, establishing the first concrete safety standards for advanced AI systems under its landmark AI Act, according to the Center for AI Safety.

The code addresses a critical gap in the EU's June 2024 AI Act, which required developers of high-risk general-purpose AI systems to "assess and mitigate possible systemic risks" but left the methodology undefined. The Act targets models trained using 10^25 FLOPs or greater—a threshold that captures frontier AI systems.

For finance leaders deploying or procuring AI systems, the code clarifies compliance expectations around transparency, copyright protection, and safety standards that were previously ambiguous. The ambiguity had created uncertainty for both developers and regulators about whether safety practices met legal requirements.

The code represents the EU's attempt to operationalize vague regulatory language into actionable guidance. Whether it succeeds depends on how strictly regulators enforce it and whether the standards prove technically feasible for AI providers.

Watch for: How major AI providers respond to the code and whether the EU issues enforcement guidance in coming months.