The G7 Hiroshima AI Process Code of Conduct (2023)

The Hiroshima AI Process was launched by the Group of Seven (G7) under Japan’s 2023 presidency, named for the city that hosted the G7 leaders’ summit that year. Its central output, published on 30 October 2023, was the Hiroshima Process International Code of Conduct for Organizations Developing Advanced AI Systems, issued alongside a companion set of International Guiding Principles. The code is one of the earliest multilateral attempts to set expectations directly for the companies building frontier AI.

The code provides voluntary guidance for organizations developing the most advanced AI systems, including foundation models and generative AI, across the full lifecycle of design, development, deployment, and use. Its actions include identifying and mitigating risks before and after deployment, publicly reporting on systems’ capabilities and limitations, investing in security controls, developing content-authentication and provenance mechanisms such as watermarking, and prioritizing research on safety and societal risks. It is framed as a living document, intended to be reviewed and updated through ongoing multistakeholder consultation.

A defining feature of the instrument is what it is not. The code is explicitly non-binding. It offers “voluntary guidance for actions by organizations” and leaves it to individual jurisdictions to decide how, or whether, to implement those actions. It creates no enforcement mechanism and imposes no legal obligation on any company. Its force comes from political endorsement by the G7 and the reputational expectation that signatory governments and participating firms will act consistently with it.

The Hiroshima Process sits within a wider 2023 wave of AI governance activity that also included the UK’s Bletchley Declaration and the US executive order on AI. Its distinctive contribution was to address obligations to the developers of advanced systems through an internationally agreed code, complementing the binding legislative route the EU pursued with its AI Act. The version cited here is the copy published in the European Commission’s digital-strategy library.