Hiroshima Process International Code of Conduct for Organizations Developing Advanced AI Systems

Summary

Voluntary code that turns the Guiding Principles into 11 actions for organisations developing advanced AI systems. They call for internal and independent external testing such as red-teaming at several points in the lifecycle, with attention to chemical, biological, radiological and nuclear risks, offensive cyber capabilities, self-replicating models and chain-reaction events; post-deployment monitoring and incident sharing; and public transparency reports. Other actions cover risk management and governance policies, security of model weights and insider threats, content authentication such as watermarking, safety research, international standards and data protection.

Safety topics

Status notes

Welcomed by the G7 Leaders in their Statement on the Hiroshima AI Process of 30 Oct 2023, which calls on organisations developing advanced AI systems to commit to applying the Code and says it will be reviewed and updated as necessary. The Code is part of the Hiroshima AI Process Comprehensive Policy Framework (G7-03). Organisations report on how they apply it through the OECD's HAIP Reporting Framework (OECD-02), launched in February 2025; G7 digital ministers commended a revised Reporting Framework in May 2026 (G7-04). The Seoul Declaration of May 2024 (INT-04) says its signatories continue to focus on supporting the operationalisation of the Code.

Sources

Open the official source https://www.mofa.go.jp/files/100573473.pdf

Secondary source: https://www.soumu.go.jp/hiroshimaaiprocess/pdf/document01_en.pdf

Local copy

Verification

History

1. 2026-10-18related: [{"slug": "hiroshima-process-guiding-principles-advanced-ai", "id": "G7-01"}, {" → [{"slug": "hiroshima-process-guiding-principles-advanced-ai", "id": "G7-01"}, {" — two-way link added: G7-03, G7-05, INT-20 link(s) to G7-022026-10-18-fix-related-links · Claude Code (claude-opus-5-5)
2. 2026-10-17notes: The OECD's Hiroshima AI Process (HAIP) Reporting Framework, launched 7 Feb 2025, → Welcomed by the G7 Leaders in their Statement on the Hiroshima AI Process of 30 — the Leaders' statement of 30 Oct 2023 and the review clause added; the revised Reporting Framework (May 2026) added; the Seoul Declaration sentence restated in its own words2026-10-17-recheck-international-1 · Claude Code (claude-opus-5-5)
3. 2026-10-17safety_topics: ["general-purpose / frontier AI", "evaluation & testing", "transparency & labell → ["general-purpose / frontier AI", "evaluation & testing", "catastrophic risk", " — seven values, above the limit of six; 'accountability & liability' dropped (the Code has no liability provision)2026-10-17-recheck-international-1 · Claude Code (claude-opus-5-5)
4. 2026-10-17summary: Voluntary guidance that turns the Guiding Principles into 11 actions for advance → Voluntary code that turns the Guiding Principles into 11 actions for organisatio — the stored text lists the specific risks testing must attend to (CBRN, offensive cyber, self-replication, chain reactions); 'testing before and after deployment' and 'weapons misuse' were imprecise2026-10-17-recheck-international-1 · Claude Code (claude-opus-5-5)
5. 2026-10-17secondary_url: https://www.soumu.go.jp/hiroshimaaiprocess/pdf/document05_en.pdf → https://www.soumu.go.jp/hiroshimaaiprocess/pdf/document01_en.pdf — the G7 Leaders' Statement on the Hiroshima AI Process (MIC Japan), the official source that dates the Code and shows the Leaders welcomed it; the old secondary URL was a second copy of the same undated text2026-10-17-recheck-international-1 · Claude Code (claude-opus-5-5)

Related records

G7 Hiroshima Guiding Principles G7 · Published (non-binding, current)

HAIP Reporting Framework OECD / G7 · Published (non-binding, current)

Frontier AI Safety Commitments AI Seoul Summit (2024) · Published (non-binding, current)

Seoul Declaration AI Seoul Summit (2024) · Published (non-binding, current)

HAIP Comprehensive Policy Framework G7 · Published (non-binding, current)

G7 Leaders' Statement on AI for Prosperity (Kananaskis, 2025) G7 · Published (non-binding, current)

HAIP Friends Group Action Plan 2026 Hiroshima AI Process Friends Group · Published (non-binding, current)

Citation

G7 Leaders (Hiroshima AI Process, Japanese G7 Presidency). Hiroshima Process International Code of Conduct for Organizations Developing Advanced AI Systems. 2023-10-30. https://www.mofa.go.jp/files/100573473.pdf. Status: Published (non-binding, current); verified 2026-10-17 (level A). Local copy archive/G7-02/2026-10-17/official.pdf retrieved 2026-10-17 14:28:17 UTC, SHA-256 1830bb44354349621fed760e31c748e1dd1ce150334c6216457750860bd130a2. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record G7-02.

Improve this record

Know of an amendment, a status change or an error? Report news or an error on GitHub; the form opens with this record filled in.