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

What it is

The Hiroshima AI Process was launched by the G7 under Japan’s presidency in 2023. It produced two documents that G7 Leaders endorsed on 30 October 2023. These are the Hiroshima Process International Guiding Principles for Organizations Developing Advanced AI Systems and the International Code of Conduct, which turns the principles into more detailed actions.

The Code is voluntary. To make it measurable, the G7 asked the OECD to build a monitoring mechanism. After a pilot in 2024, the OECD launched the HAIP Reporting Framework on 7 February 2025, a public questionnaire through which organisations disclose how they manage AI risk. Version 2.0 followed on 28 May 2026.

Who it applies to

The Code is addressed to organisations developing the most advanced AI systems, including foundation models and generative AI. Academia, civil society and the public sector are also invited to apply it.

Reporting has widened. The first reporting round drew submissions from large model developers and technology companies. HAIP 2.0 distinguishes model developers, application developers and deployers, so each can answer the questions relevant to its role, and it was designed to make participation easier for small and medium enterprises. More than 50 organisations have pledged to report.

What it requires

The 11 actions

1. Take appropriate measures throughout development to identify, evaluate and mitigate risks across the AI lifecycle, including red-teaming and testing before deployment.
2. Identify and mitigate vulnerabilities, incidents and patterns of misuse after deployment, including placement on the market.
3. Publicly report advanced AI systems’ capabilities, limitations and domains of appropriate and inappropriate use, to support transparency and accountability.
4. Work towards responsible information sharing and reporting of incidents among organisations developing advanced AI, including with industry, governments, civil society and academia.
5. Develop, implement and disclose AI governance and risk management policies, grounded in a risk-based approach, including privacy policies and mitigation measures.
6. Invest in and implement robust security controls, including physical security, cybersecurity and insider threat safeguards across the lifecycle.
7. Develop and deploy reliable content authentication and provenance mechanisms, such as watermarking, so users can identify AI-generated content where technically feasible.
8. Prioritise research to mitigate societal, safety and security risks, and invest in effective mitigation measures.
9. Prioritise developing advanced AI systems to address the world’s greatest challenges, such as the climate crisis, global health and education.
10. Advance the development and, where appropriate, adoption of international technical standards.
11. Implement appropriate data input measures and protections for personal data and intellectual property.

Reporting

The original HAIP Reporting Framework asks organisations to describe their practice across seven areas.

• Risk identification and evaluation
• Risk management and information security
• Transparency reporting on advanced AI systems
• Organisational governance, incident management and transparency
• Content authentication and provenance
• Research and investment to advance AI safety and mitigate societal risks
• Advancing human and global interests

HAIP 2.0 keeps alignment with the Code, simplifies and structures the questionnaire, tailors questions by role, links to the OECD.AI Catalogue of Tools and Metrics for Trustworthy AI so organisations can cite the tools they use, and covers newer capabilities such as agentic AI. Reports are published on OECD.AI, submissions are accepted on a rolling basis, and organisations are invited to update them annually.

Key dates

G7 Leaders endorsed the Code and Guiding Principles on 30 October 2023. The OECD launched the HAIP Reporting Framework on 7 February 2025 and published the first round of reports later that year. HAIP 2.0 launched on 28 May 2026 in Paris, at the G7 Digital and Tech Ministerial Meeting under the French presidency.

How it relates to other frameworks

The Code builds on the OECD AI Principles and translates them into commitments for developers of advanced AI. It covers similar ground to the EU General-Purpose AI Code of Practice, which is the more detailed and legally anchored instrument for general-purpose model providers in the EU, and the OECD has published analysis on how HAIP 2.0 supports interoperability with the EU AI Act.

For deployers, the Code is most useful in two ways. A vendor’s HAIP report is a public, structured source for AI vendor due diligence. And under HAIP 2.0, a deployer can report its own practices against a recognised international framework, drawing on the governance policies, incident handling and risk management it already runs under ISO/IEC 42001 or the NIST AI RMF.