The UK AI Safety Institute (AISI, renamed AI Security Institute in early 2025) operates within the UK's Department for Science, Innovation and Technology and was the first state-backed organisation dedicated to evaluating advanced AI systems. It runs pre-deployment evaluations of frontier models — through agreements with developers including OpenAI, Anthropic, Google DeepMind, and Meta — on capability, bio/chemical/cyber misuse risk, and societal-impact dimensions.
AISI publishes evaluation methodology and selected results, contributes to international evaluation standards, and serves as a working model for state-led AI assessment that other governments (US AISI, EU AI Office) have followed.