Hiroshima Process International Code of Conduct for Advanced AI Systems
Eleven voluntary actions for frontier AI developers covering red-teaming, incident reporting, security controls and content authentication. Now backed by an OECD reporting framework that makes adherence comparable.
Key dates
- Oct 30, 2023Endorsed by G7 leaders
- Feb 7, 2025OECD reporting framework launched
Risk areas addressed
Who it applies to
Organisations developing the most advanced AI systems, including foundation and generative models.
Penalties & enforcement
None (voluntary framework).
Enforced by: None (voluntary, with OECD-hosted transparency reporting)
G7 Hiroshima Code of Conduct carries no direct penalty today, so its value is evidential: teams adopt it to show a named standard behind their controls for safety & robustness, transparency and security. Mapping an existing AI inventory against it is usually enough to surface the gaps, and that same evidence tends to carry over to the binding regimes that follow.