UK AI Security Institute
The AI Security Institute (AISI) equips governments with a scientific understanding of the risks posed by advanced AI. It tests leading AI systems, researches risk mitigations, informs policymakers and works with developers, researchers and allied governments to keep the public safe.
At a Glance
- UK central government and national-security stakeholders
- Allied governments and international AI-safety institutes
- Frontier AI developers
- Universities and academic researchers
- +2 more
AI Tools by UK AI Security Institute
(2)Inspect Evals
LLM Benchmark and Safety Evals
Inspect AI
Open Source LLM Eval Framework
Discussions
No discussions yet
Be the first to start a discussion about UK AI Security Institute
Latest News
Pretraining data filtering in open-weight biological models
Can AI coding agents rebuild biological AI tools from a paper?
Investigating alignment failures in AISI's incident
Towards technical best practice for frontier AI development
Products & Services
An open-source AI evaluation platform for running evaluations quickly and easily; it is intended for AISI, governments, AI developers and the wider research community.
An open collection of evaluations for Inspect AI, including dozens of tests that make AI safety research more accessible. The supplied repository is published by the UKGovernmentBEIS GitHub organisation under an MIT license.
An open evaluation tool in AISI's Inspect ecosystem, listed by AISI as one of its government-backed evaluation tools.
An open tool for cyber evaluations, listed by AISI among the evaluation tools used by governments, companies and academics.
Market Position
AISI positions itself as the world's first state-backed organisation dedicated to advanced-AI safety and security, combining government authority, independent technical evaluations and an agile startup-like operating model. Its closest comparators are the US AI Safety Institute/Center for AI Standards and Innovation and other national AI safety institutes; unlike commercial AI labs, AISI's role is independent public-interest testing, government preparedness and open scientific infrastructure rather than selling models or AI products.
Leadership
Founders
Ian Hogarth
Chair of AISI; Co-founder of Songkick and venture capitalist (Plural); appointed to lead the Frontier AI Taskforce in 2023.
Rishi Sunak
Former UK Prime Minister who spearheaded the creation of the Frontier AI Taskforce and the AI Safety Institute at Bletchley Park.
Executive Team
Henry de Zoete OBE
Director
Former Prime Minister's AI Advisor and UK Government Investment and AI Adviser; also a technology investor and founder.
Jade Leung
Chief Technology Officer; Prime Minister's AI Advisor
Previously led the Governance team at OpenAI.
Board of Directors
Founding Story
AISI was launched at the UK-hosted AI Safety Summit in November 2023 as the permanent successor to the Frontier AI Taskforce, which had been announced in April 2023. The initial vision was to place rigorous technical expertise at the heart of government: evaluate frontier systems, conduct foundational safety research, share evidence internationally and help governments respond to fast-moving AI risks. The organisation later adopted the name AI Security Institute on 14 February 2025.
Business Model
Revenue Model
AISI is a publicly funded government research organisation rather than a commercial company. Its work is funded through UK government appropriations and long-term DSIT resourcing, with additional grant programmes and collaborations; its open-source tools are made available without a commercial subscription model.
Target Markets
- UK central government and national-security stakeholders
- Allied governments and international AI-safety institutes
- Frontier AI developers
- Universities and academic researchers
- AI safety, security and alignment researchers
- Policymakers, regulators and civil society
- Government preparedness and policymaking for advanced-AI risks
- Independent pre-release testing of frontier models
- Cybersecurity and cyber-misuse evaluation
- Chemical and biological misuse evaluation
- Testing safeguards and jailbreak resistance
- Alignment and AI-control research
- UK government departments and national-security partners
- Google DeepMind
- OpenAI
- Anthropic