Timeline

The G7 agrees a Hiroshima code of conduct for AI

Eleven non-binding principles for advanced-AI developers, carrying no legal force in any G7 jurisdiction, published the same day as the US executive order.

  • Government & policy
  • Minor

G7 leaders published the Hiroshima Process International Code of Conduct for Organizations Developing Advanced AI Systems, alongside a companion set of guiding principles, closing out the process they had launched at the Hiroshima Summit that May. The code set out eleven actions the G7 asked developers of the most advanced foundation models and generative systems to follow — among them identifying and mitigating risks throughout a model’s lifecycle, publicly reporting capabilities and limitations, investing in security controls, and prioritising research into risks such as bias, misinformation and misuse.

The code was explicitly voluntary. It created no legal obligation in any G7 member state, no enforcement mechanism, and no body to monitor compliance; it functioned as a statement of shared expectations that governments hoped developers would adopt on their own or that could later inform binding rules developed separately in each jurisdiction. The document arrived the same day the White House issued a domestic instrument with actual regulatory teeth — Joe Biden’s executive order on AI, which imposed mandatory safety-testing reporting requirements on developers of the largest models. The contrast underlined what the G7 process could and could not deliver: political agreement across seven governments with very different domestic AI industries and regulatory traditions was reachable only at the level of shared principles, not shared law.

The Hiroshima code became one of several voluntary international frameworks published in quick succession in late 2023 — alongside commitments major AI companies made directly to the White House earlier that year, and the Bletchley Declaration signed at the UK’s AI Safety Summit two days later — that together signalled international convergence on the broad shape of AI governance without any single instrument binding developers across borders. Its practical effect depended entirely on whether individual G7 governments translated its principles into domestic law, which happened unevenly and slowly over the following two years.