The G7 Hiroshima
AI Process, Explained.

Named for the city where the G7 leaders met in 2023, the Hiroshima AI Process was an attempt by the world's leading industrial democracies — the US, UK, France, Germany, Italy, Canada, Japan, and the EU — to set shared expectations for advanced AI while binding regulation was still years away. Its outputs, delivered later that year, were the International Guiding Principles for Advanced AI Systems and a voluntary International Code of Conduct for organisations developing them.

What the Hiroshima Process produced

The Guiding Principles set out eleven expectations for organisations developing advanced AI: identifying and mitigating risks across the lifecycle, testing systems before and after deployment, being transparent about capabilities and limitations, sharing information among developers and with governments, investing in security controls, and developing mechanisms to identify AI-generated content, among others. The Code of Conduct translated these principles into more operational guidance for developers to follow voluntarily.

This was, in late 2023, the most detailed internationally agreed statement of what responsible frontier AI development should involve. Unlike a bare declaration, it descended to the level of specific practices, giving companies and regulators a common reference point for what good behaviour looks like.

The strength of the club model

The G7 approach has a real advantage: it is fast and coherent. A small group of aligned democracies that already cooperate closely can reach agreement far more quickly than a universal negotiation, and can set standards that carry weight because their members host much of the frontier industry. Governance that starts in a like-minded club can later be widened — the pattern of many international regimes, where a core group sets the standard and others converge on it over time.

The Hiroshima outputs also fed into the wider ecosystem, informing the OECD's work and aligning with the direction taken at Bletchley and Seoul. As a way of establishing early norms among the states that matter most for AI development, the club model is efficient and underrated.

The ceiling of the club model

• It excludes China. The most significant AI developer outside the G7 is not in the room. A code of conduct that binds only democracies does not constrain the full field of frontier development, and cannot become universal governance on its own.
• It is voluntary. Like the Seoul commitments, the code has no verification and no enforcement. It describes good practice; it does not compel it.
• It is a floor, not a ceiling. Guiding principles set minimum expectations, but the hardest questions — binding thresholds, independent monitoring, consequences for violation — are left to future, harder processes.

A club of democracies can write the rulebook quickly, and that is worth doing. But a rulebook that the other main player never signs, with no one checking and no penalty for ignoring it, is a first draft of governance — not the finished document.

Naoto Nakada, Founder · Nakada Foundation to Save Humanity

Where it fits in the larger picture

The Hiroshima Process is best understood as one of several parallel tracks — alongside the summit series, the OECD, and the UN — through which early AI norms are being assembled. Its contribution is a detailed, agreed statement of responsible practice among the leading democracies, achieved quickly by leveraging existing trust. Its limit is that a like-minded club cannot deliver universal, binding, verified governance by itself, because the actors it excludes are precisely the ones a comprehensive regime must include. The realistic role of the G7 model is to set the standard and build the momentum that a broader, binding agreement can later formalise — a starting point that only matters if it is eventually widened and given teeth.