G7 Hiroshima Process Code of Conduct
// Voluntary Code - Frontier Discipline Before Law
When the world's leading economies wanted discipline on frontier AI without waiting for treaties, they wrote a code of conduct. It is voluntary - and it is becoming the template that binding general-purpose AI rules copy, which makes early alignment cheap insurance.
Key Highlights
What the code asks of advanced-AI organizations.
- 01
The Hiroshima Process Code of Conduct for Organizations Developing Advanced AI Systems was agreed by G7 leaders in October 2023 as the operational arm of the G7's AI vision.
- 02
The code sets voluntary commitments across risk identification and evaluation, mitigation, incident disclosure, transparency reporting, cybersecurity investment, and content provenance research.
- 03
It targets organizations developing the most capable frontier systems - but its practices are already migrating into binding rules, notably the EU's GPAI Code of Practice which operationalizes similar duties for EU law.
- 04
Reporting moved from paper to practice: an OECD-run pilot has been monitoring application of the code since July 2024, normalizing structured disclosure for advanced-AI developers.
- 05
For adopters rather than developers, the code is a vendor-quality checklist: suppliers who follow it can evidence evaluation, security, and provenance discipline.
Action Items for Governance and Compliance
Four moves that make the code your advantage rather than another acronym.
Publish transparency reports for advanced capabilities
A01Name a reporting owner with executive sponsorship and a fixed publication cadence.
Document capability evaluations, misuse mitigations, and incident disclosures in a consistent public format.
Invest in content provenance
A02Sponsor provenance and watermarking research as a product requirement, not a lab experiment.
Measure synthetic-content marking coverage so claims about provenance survive scrutiny.
Strengthen adversarial testing and security
A03Fund a recurring red-team cadence with direct executive reporting lines.
Retain penetration and abuse-testing records as good-practice evidence regulators increasingly expect.
Use the code as a supplier checklist
A04Add code-alignment questions to AI vendor assessments during procurement.
Prefer suppliers who can evidence evaluation, incident, and provenance practices over those who cannot.
Engineering Translation
High-level capabilities that turn code commitments into artifacts.
>_Evaluation reports generated continuously from live harnesses instead of written before launches.
>_Incident disclosures drafted from structured event data, ready for publication review.
>_Provenance coverage tracked as a product metric visible on engineering dashboards.
Delivered as capabilities, not paperwork
Each obligation above is translated into product-level engineering capabilities. The underlying stack and internal tooling are intentionally abstracted here; they are covered in technical briefings.
Educational summary only. This is not legal advice. Confirm obligations with qualified counsel for your jurisdiction.
Meet Tomorrow's GPAI Rules Today
Book a technical briefing to wire evaluation, disclosure, and provenance capabilities directly into your platform.