OECD AI Principles, G7 Hiroshima Code and Seoul Commitments: international voluntary AI frameworks

Scope of this explainer

This explainer draws on three sources: the European Commission's page on the Hiroshima Process International Code of Conduct, the Seoul Declaration of 21 May 2024 with its annex, and the page for the International AI Safety Report 2025. The OECD AI Principles and the Frontier AI Safety Commitments appear in these sources only as references. The explainer describes them only as the documents do.

What the Hiroshima Code asks of developers

The Commission's page, dated 30 October 2023, says the code is voluntary. It states that the International Code of Conduct for Organizations Developing Advanced AI Systems aims to promote safe, secure, and trustworthy AI worldwide and will provide voluntary guidance for actions by organizations developing the most advanced AI systems.

A press release listed on the same page, also dated 30 October 2023, says the Commission welcomes the G7 leaders' agreement on guiding principles and a voluntary code for AI developers. It places both under the Hiroshima AI process.

The page sets two conditions on how the actions are applied. First, it says organizations should follow the actions in line with a risk-based approach. Second, it says the actions apply across the lifecycle, covering design, development, deployment and use of advanced AI systems, when and as applicable.

The page describes the actions as a non-exhaustive list. It calls the list a living document that builds on the existing OECD AI Principles. The page offers the code as a download. The text of the individual actions sits in that download and is not reproduced on the page itself.

What the Seoul Declaration asks

The Seoul Declaration says leaders gathered at the AI Seoul Summit on 21 May 2024 and affirmed a common dedication to international cooperation and dialogue on AI. The declaration names the leaders as representing Australia, Canada, the European Union, France, Germany, Italy, Japan, the Republic of Korea, the Republic of Singapore, the United Kingdom, and the United States of America.

The declaration ties the Seoul process back to the Hiroshima Code. It says the leaders continue to focus on supporting the operationalisation of the code. It also says the leaders recognize the importance of interoperability between AI governance frameworks in line with a risk-based approach.

Where the Frontier AI Safety Commitments appear

The declaration refers to the Frontier AI Safety Commitments in one sentence. The leaders recognize the particular responsibility of organizations developing and deploying frontier AI, and in that connection note the commitments. The declaration text does not list what the commitments contain, so the contents cannot be described from the documents supplied.

The annex, the Seoul Statement of Intent toward International Cooperation on AI Safety Science, is addressed to governments. It is not addressed to developers. It lists steps the signatories say may be taken. These include sharing information about models, including their capabilities, limitations and risks as appropriate. They also include monitoring AI harms and safety incidents, and exchanging or jointly creating evaluations, data sets and associated criteria where appropriate. The list also includes promoting appropriate research security practices in the field.

The annex adds that the signatories intend to promote the development and adoption of international standards. It pairs this with accelerating evidence-based approaches to AI safety.

How the voluntary frameworks relate to binding AI laws

Neither the code page nor the declaration creates a legal duty in the text supplied. The Commission page says different jurisdictions may take their own unique approaches to implementing the actions in different ways.

The annex treats alignment as a matter for each signatory's own policy. It says the signatories intend to use assessments such as the International AI Safety Report to guide and align their respective policies where appropriate, in line with their governance frameworks.

The International AI Safety Report adds a view on regulators. It says there are various technical methods to assess and reduce risks from general-purpose AI that developers can employ and regulators can require, but all have limitations. The report describes itself as scientific input and says it does not recommend specific policies.

How they relate to security programs

None of the supplied documents sets out a security control or a compliance step for a company. The material that touches security comes in three forms.

The first is incident monitoring and research security. The annex names the monitoring of AI harms and safety incidents and the promotion of research security practices as possible steps for the signatories. The second is the International AI Safety Report, a source of evidence. The report page says it is the first comprehensive review of scientific research on the capabilities and risks of general-purpose AI systems, published in January 2025. On cyber offence, the report says state-sponsored actors are actively exploring AI to survey target systems.

The third is risk management frameworks. The report says companies and governments are developing early warning systems and risk management frameworks. It says some trigger specific mitigation measures when there is new evidence of risks. Others require developers to provide evidence of safety before releasing a new model.

The code is the voluntary reference point for developers of advanced AI systems. The Seoul documents connect it to the OECD AI Principles and to government-level cooperation on safety science. Binding obligations, where they exist, come from each jurisdiction's own approach, and the supplied documents do not name them.