Anthropic to sign the EU Code of Practice

Source: Anthropic Date: 2025-07-21 URL: https://www.anthropic.com/news/eu-code-practice

Summary

Anthropic announced intent to sign the EU’s General-Purpose AI Code of Practice, which establishes mandatory Safety and Security Frameworks for frontier AI developers. Key commitment: CBRN risk evaluations. Framing: EU AI Act compliance compatible with innovation; AI could add 1T+ euros annually to EU economy by mid-2030s. Anthropic anchored the commitment to its existing RSP and Frontier Model Forum participation.

Implications

• Policy / EU AI Act compliance as market access. The EU AI Act’s GPAI provisions apply to frontier models distributed in the EU — signing the Code of Practice is the compliance mechanism for continued EU market access. Anthropic’s framing (compatible with innovation) is the standard approach to European AI regulation by US companies.
• CBRN evaluation commitment. Explicitly committing to CBRN risk evaluations in the EU Code context links to ASL-3 activation (May 2025) — Anthropic is taking its existing ASL-3 evaluation framework and presenting it as Code of Practice compliance. The cost of compliance is low because the infrastructure already exists.
• RSP as the compliance template. “Anchoring to existing RSP” is Anthropic saying its existing safety framework meets the EU Code’s requirements without new work. This is an argument against regulatory fragmentation — one framework, multiple jurisdictions.
• Timing with US energy advocacy. The EU Code signing (July 21) happened the same day as the “Build AI in America” energy report — Anthropic was simultaneously engaging European regulators and US policymakers on the same day, managing two different regulatory relationships.
• Watch: whether Anthropic formally signed the Code and when; how the EU AI Act enforcement mechanism interacted with the voluntary Code of Practice commitments; whether CBRN evaluations under the Code produced findings inconsistent with ASL-3 assessments.