The UK AI Safety Institute (AISI) was established in November 2023, around the time of the AI Safety Summit at Bletchley Park, and evolved out of the UK’s Frontier AI Taskforce that had been announced in April 2023. The UK government describes it as “the first state-backed organisation focused on advanced AI safety for the public interest,” making it the prototype for a wave of national AI safety institutes that followed in the United States, the European Union, and elsewhere.
The institute’s stated mission is to “minimise surprise to the UK and humanity from rapid and unexpected advances in AI.” It pursues this through three main activities: conducting evaluations of cutting-edge AI systems to measure safety-relevant capabilities, carrying out foundational research on AI risk, and facilitating information exchange among policymakers, companies, academia, and the public. A guiding idea, in the institute’s own words, is to move the discussion of AI risk “from the speculative and philosophical, further towards the scientific and empirical.”
The practical contribution of AISI is hands-on testing capacity that governments had previously lacked. Rather than relying solely on the assurances of the companies building frontier models, the institute set out to build the technical infrastructure to evaluate models directly, including red-teaming and capability assessments for dangerous uses. This positioned a public body to act as an independent check, and the institute went on to sign agreements with leading labs to access models for pre-deployment testing.
It is worth being clear about what the institute is and is not. AISI is a research and evaluation body, not a regulator; it has no statutory power to license, approve, or ban AI systems, and its evaluations inform government and the public rather than gate products by law. Its influence is evidential and reputational. Even so, by becoming the first government to stand up a dedicated technical safety institute, the UK established a template that several other jurisdictions copied, and these institutes began to coordinate as an international network at the subsequent Seoul and Paris summits.