The AI Safety Summit at Bletchley Park was an international gathering held on November 1 and 2, 2023, at Bletchley Park in the United Kingdom, the historic site of Britain's Second World War codebreaking effort, convened by the UK government to coordinate international policy on frontier AI risk. Representatives from 28 countries and the European Union, alongside major AI companies and researchers, attended, producing the Bletchley Declaration, a joint statement acknowledging the risks of advanced AI and committing to international cooperation.
The declaration
The Bletchley Declaration was signed by all attending governments, including the United States, China, and the European Union, an unusual instance of geopolitical rivals agreeing on shared language about AI risk. It stated that most AI risks arise from unintended issues of control as well as potential misuse, and specifically flagged risks associated with frontier AI capable of causing serious harm, whether deliberately or unintentionally, echoing concerns raised earlier that year in the Pause Giant AI Experiments letter. The declaration stopped short of proposing binding rules, focusing instead on shared scientific understanding and future cooperation.
Participants and side outcomes
Executives and researchers from leading labs attended, including representatives from OpenAI, Google DeepMind, and Anthropic, alongside academics and safety researchers such as Yoshua Bengio, who was tasked with leading an international "State of AI" scientific report modeled loosely on climate science assessments. The summit is closely associated with the announcement of national AI safety institutes, including the UK AI Safety Institute and the US AI Safety Institute, bodies intended to conduct evaluations of frontier models before and after release, connecting to broader efforts around red teaming and evaluation of advanced systems.
Legacy
The Bletchley summit was the first in a planned series of AI safety summits, followed by a South Korean summit in 2024 and a French summit in 2025, reflecting an attempt to institutionalize recurring international coordination on AI risk in a manner comparable to climate diplomacy. Its practical effect was mostly declaratory rather than regulatory, in contrast to binding measures such as the EU AI Act, but it is frequently cited as marking the point at which frontier AI risk became a mainstream matter of head-of-state-level diplomacy rather than a niche research concern, feeding into ongoing debates about AI governance and the role of existential risk arguments in shaping policy.