UK AI Safety Institute

The UK stood up the world's first government AI safety institute in November 2023. Its structure, scope, and access model are templates other nations are following.

34 min · Reviewed 2026

From Bletchley to Institute

The UK hosted the first global AI Safety Summit at Bletchley Park in November 2023. Coming out of it, Prime Minister Rishi Sunak announced the AI Safety Institute (AISI), a government body tasked with evaluating frontier AI. It began operating immediately with a direct mandate from the Department for Science, Innovation and Technology.

What AISI does

Pre-deployment evaluations of frontier models from OpenAI, Anthropic, Google DeepMind, Meta, and others
Research on misuse risks — cyber, bio, autonomous systems, deepfakes
Public publication of evaluation results and methodologies
Coordination with the US AI Safety Institute (later CAISI) and similar bodies
Grants to external researchers and open-source safety tools

Staffing and output

Employs dozens of researchers including alumni from DeepMind, OpenAI, Anthropic
Led initially by Ian Hogarth and Jade Leung, with Geoffrey Irving as research director
Published detailed evaluation reports on Claude, GPT-4o, o1, Llama, and others throughout 2024-2025
Identifies failure modes that inform labs' own safety work

We are not a regulator. We are a source of evidence. What governments do with that evidence is up to them.
— Early AISI leadership, paraphrased

The big idea: AISI showed that a government can build technical AI evaluation capacity without new laws. Every serious AI policy actor now has something like it or is building one.

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-safety2-uk-aisi-creators

What made the UK AI Safety Institute historically significant when it was announced in November 2023?
1. It was the first AI lab established at Bletchley Park
2. It was the first institute to publish AI research papers
3. It was the first AI company to receive government funding
4. It was the world's first government AI safety institute
Under which government department does the UK AI Safety Institute operate?
1. Department of Health and Social Care
2. Department for Science, Innovation and Technology
3. Ministry of Defence
4. Treasury
What type of evaluations does the UK AI Safety Institute conduct on frontier AI models?
1. Post-market surveillance evaluations
2. Retrospective incident analysis evaluations
3. Real-time continuous monitoring evaluations
4. Pre-deployment evaluations
Which of the following was listed as an area of misuse research conducted by AISI?
1. Entertainment piracy risks
2. Retail fraud risks
3. Agricultural automation risks
4. Biological weapons risks
What fundamental limitation did the lesson describe regarding AISI's authority over AI companies?
1. It has no ability to hire researchers
2. It has no statutory power to compel companies to submit models
3. It is forbidden from publishing evaluation results
4. It lacks access to government funding
How do major AI labs like OpenAI and Anthropic participate in AISI's evaluation process?
1. Through competitive bidding processes
2. Through legally binding contracts with the UK government
3. Through automatic mandatory submission requirements
4. Through voluntary commitments made at international summits
In early 2025, what change did the UK make to the AI Safety Institute?
1. Converted it into a private company
2. Merged it with the US AI Safety Institute
3. Renamed it the AI Security Institute
4. Dissolved it and created a new agency
What was the main shift in focus after the UK renamed its AI institute in early 2025?
1. Toward entertainment industry concerns
2. Toward national-security-relevant risks
3. Toward commercial AI applications
4. Toward consumer privacy protection
Who was mentioned in the lesson as serving as AISI's research director?
1. Demis Hassabis
2. Dario Amodei
3. Geoffrey Irving
4. Sam Altman
What did the lesson quote AISI as stating about its own role?
1. We are the world's foremost AI research lab.
2. We are not a regulator. We are a source of evidence.
3. We are the enforcement arm of UK AI law.
4. We are bound by international AI treaties.
Which AI models were specifically mentioned as having been evaluated and reported on by AISI in 2024-2025?
1. Siri, Alexa, Cortana, and Bixby
2. Claude, GPT-4o, o1, and Llama
3. DALL-E, Midjourney, Stable Diffusion, and Claude
4. Watson, Deep Blue, AlphaGo, and AlphaFold
What types of grants does the lesson say AISI provides?
1. Grants to external researchers and open-source safety tools
2. Grants to AI companies for product development
3. Grants to universities for general computer science education
4. Grants to startups for commercial AI applications
At what location was the first global AI Safety Summit held, where AISI was announced?
1. Bletchley Park
2. Davos
3. San Francisco
4. Geneva
What was identified in the lesson as the 'big idea' regarding AISI's significance for AI policy?
1. That AI development should be paused indefinitely
2. That AI companies should be nationalized
3. That international treaties are required for AI safety
4. That a government can build technical AI evaluation capacity without new laws
With which international body did the lesson say AISI coordinates its work?
1. The United Nations Educational body
2. NATO's cyber defense division
3. The European Union's AI Office
4. The US AI Safety Institute (later CAISI)

← Back to interactive lesson

Tendril · Creators · Ethics & Society