Twenty-eight governments meet at Bletchley Park and agree, for the first time, on what they fear

On the morning of 1 November 2023, twenty-eight national delegations and representatives from the European Commission gathered in the Victorian rooms of Bletchley Park, the codebreaking site that had hosted Alan Turing's wartime cryptanalysis work. The choice of venue was deliberate. Prime Minister Rishi Sunak had spent eighteen months positioning the United Kingdom as the natural neutral ground for an international conversation about advanced AI, and Bletchley supplied the iconography.

The substantive output of the two-day summit was the Bletchley Declaration, a one-page document signed by the United States, China, the United Kingdom, the European Union, France, Germany, India, Brazil, Japan and nineteen others. The text committed signatories to nothing binding. It named, for the first time in a multilateral document, a class of "frontier AI" models whose capabilities could exceed the most advanced existing systems and whose risks therefore deserved coordinated attention. As reported by the BBC and the Financial Times, the inclusion of China alongside the US and EU was the diplomatic surprise of the week.

What was actually agreed

Behind the declaration, two more concrete steps were quietly negotiated. The first was a state-led testing regime for the next generation of frontier models. OpenAI, Anthropic, Meta, Google DeepMind, Microsoft, Amazon, Inflection and Mistral all agreed in principle, the UK government announced, to give national AI Safety Institutes early access to their pre-deployment models for evaluation. The second was a recurring summit cadence. Korea would host the next meeting in May 2024, France the one after that. The institution-building had begun.

Nothing legally binding was signed. Something behaviourally binding was.

Critics on both flanks

The summit drew criticism from two directions. Civil society groups, including the Ada Lovelace Institute and the Mozilla Foundation, noted that the agenda was dominated by long-horizon "frontier" risk and largely ignored the present-day harms of bias, surveillance and labour displacement. The Tony Blair Institute, in commentary published the following week, took the opposite view, calling the declaration insufficiently prescriptive given the pace of capability progress and the regulatory vacuum in most of the signatory countries.

What the summit did establish is a category. Twelve months earlier, no government had been able to use the words "frontier AI" with a stable definition. By the time the Bletchley delegations went home, that phrase, and the implicit policy structure beneath it, had a multilateral foothold. The Korean summit six months later moved that foothold a few inches further.