Britain founds the AI Safety Institute
On 2 November 2023, during the Bletchley Park summit, the UK government published the command paper 'Introducing the AI Safety Institute': the Frontier AI Taskforce became a permanent institute that day. It was to develop and conduct evaluations of advanced AI systems, drive foundational safety research and facilitate information exchange.
Why it matters
A state set up a permanent body whose work is to test private laboratories' models itself rather than regulate them: the paper says in so many words that the Institute is not a regulator. Two and a half years later it is this body that independently evaluates the cyber capabilities of Claude Mythos Preview.
Evaluations were to cover the capabilities most relevant to misuse, capabilities that could worsen societal harms such as manipulation, bias and effects on democracy, and the safety of the systems and their safeguards. The Taskforce, announced in April 2023, became the Institute's core, and Ian Hogarth stayed on as chair. Money: the paper ties the initial GBP 100m to the Taskforce and promises the Institute the Taskforce's 2024/25 funding as an annual amount for the rest of the decade, subject to continued need; the record claims no separate budget for the Institute. On 10 May 2024 the Institute made its Inspect evaluations platform available to all; the documents read do not say it was its first publication. The Institute was later renamed the AI Security Institute, the name under which the atlas carries it; this pass did not read the date of the renaming.