The Bletchley Declaration at the first AI Safety Summit

The UK convened the first international AI Safety Summit at Bletchley Park, the wartime codebreaking site. Twenty-eight countries and the European Union signed the Bletchley Declaration, agreeing that frontier AI models carried potential for serious, even catastrophic, harm and that international cooperation was needed to address it.

The signatures that mattered were the US and China appearing on the same document. Cooperation between them on any technology question was scarce at that point — the US had tightened chip export controls three weeks earlier — and getting both to endorse a shared statement of risk was the summit’s principal diplomatic achievement.

The declaration contained no binding commitments. It established a direction of travel: an agreed agenda of identifying risks of shared concern and building scientific understanding of them, plus a commitment to reconvene. Separately, the UK announced the AI Safety Institute, the first state-funded body dedicated to evaluating frontier models, and several major labs agreed to give it pre-deployment access to their systems. The US announced its counterpart the same week.

The summit’s other output was a commissioned scientific assessment, chaired by Yoshua Bengio, which became the International AI Safety Report — an IPCC-style synthesis intended to give governments a shared evidence base.

The series continued in Seoul in May 2024, where sixteen companies signed Frontier AI Safety Commitments, and in Paris in February 2025, which was renamed the AI Action Summit and pivoted sharply toward opportunity and investment. The US and UK both declined to sign the Paris statement. That trajectory — from shared alarm to competitive urgency in fifteen months — is one of the clearer signals of how the policy climate changed.