UK government research organization that tests frontier AI systems, advances AI safety science, and informs policymakers about the risks and capabilities of advanced AI.
UK government research organization that tests frontier AI systems, advances AI safety science, and informs policymakers about the risks and capabilities of advanced AI.
People
Updated 05/18/26 · By grantmaking.aiInterim Director
Research Director
Chief Scientist
Chair
Funding Details
Updated 05/18/26 · By grantmaking.ai- $66,000,000
Org Details
Updated 05/18/26 · By grantmaking.aiThe AI Security Institute (AISI) was established on November 2, 2023, evolving from the UK's Frontier AI Taskforce (which had been created in April 2023) and announced at the Bletchley Park AI Safety Summit. It sits within the UK Department for Science, Innovation and Technology and was originally named the AI Safety Institute before being renamed the AI Security Institute in February 2025, reflecting an expanded focus that includes AI-enabled crime and security threats alongside its core safety research mandate.
AISI's mission is to build the world's leading understanding of advanced AI risks and solutions, to inform governments so they can keep the public safe. The organization designs and runs evaluations of frontier AI systems across domains critical to national security and public safety, including cyber misuse, biosecurity, safeguards, alignment, control, autonomy, human influence, and societal resilience. By 2025, AISI had tested more than 30 of the world's most advanced models.
The institute conducts original technical research, publishes open-source evaluation tools including Inspect, InspectSandbox, InspectCyber, and ControlArena, and releases major reports such as the inaugural Frontier AI Trends Report. In 2025 it published its first peer-reviewed study in the journal Science, covering a 76,000-participant experiment on AI-enabled persuasion.
AISI runs several major grant programs to fund external AI safety research. The Alignment Project, launched in 2025, is one of the largest global alignment research efforts and awarded over £27 million to more than 60 projects in its first round, with applications expected to reopen in summer 2026. The Systemic Safety Grants program has disbursed £8 million and a Challenge Fund has allocated £5 million.
The institute is led by Interim Director Adam Beaumont (former Chief AI Officer at GCHQ) and Chief Technology Officer Jade Leung (also the Prime Minister's AI Advisor, previously at OpenAI). Chief Scientist Geoffrey Irving and Research Director Chris Summerfield bring experience from OpenAI, Google DeepMind, and the University of Oxford. Chair Ian Hogarth is a tech investor and entrepreneur. The advisory board includes machine learning luminaries such as Yoshua Bengio. The technical team numbers more than 100 staff. AISI receives £66 million in annual government funding and has priority access to over £1.5 billion of UK compute resources.
AISI helped launch and leads the International Network for Advanced AI Measurement, Evaluation and Science, a coalition of allied-nation AI safety institutes. It maintains deep partnerships with leading AI developers including Anthropic, OpenAI, Google DeepMind, and Cohere, which provide pre-deployment model access for safety evaluations.
Theory of Impact
Updated 05/18/26 · By grantmaking.aiAISI holds that the primary bottleneck to safe AI development is a lack of rigorous, independent scientific understanding of what frontier AI systems can actually do — especially in high-stakes domains like biosecurity, cyber offense, and persuasion. By building credible evaluation infrastructure and running transparent assessments, AISI gives policymakers the evidence base they need to regulate effectively and gives developers incentive to prioritize safety. Simultaneously, AISI funds alignment and systemic safety research to close the gap between rapidly advancing capabilities and the mitigations available to address them. International coordination through the network of AI safety institutes ensures that safety standards become global rather than fragmented across jurisdictions, reducing the risk that dangerous capabilities go undetected or that safety races to the bottom.
I've spoken to several current and former UK AISI employees at various AI safety events and conferences. I think that it's the best place to work on safety in government, the team is very strong and their research directions are promising.