The Foundation Model Taskforce was the UK government’s April 2023 vehicle for AI capability-building; it was renamed the Frontier AI Taskforce in September 2023, became the AI Safety Institute in November 2023, and was renamed again as the AI Security Institute in February 2025. Each rename shifted both remit and budget line, and each shift changed who research funders and universities need to engage with on AI policy.
The Foundation Model Taskforce is the name the Department for Science, Innovation and Technology (DSIT) gave, in April 2023, to a government unit tasked with building UK capability in large-scale AI systems and evaluating their risks. It no longer exists under that name — but its institutional descendant, the AI Security Institute, still shapes the funding and evaluation landscape that UK research offices operate in today.
- What was the UK’s Foundation Model Taskforce?
- Why did it become the Frontier AI Taskforce, then the AI Safety Institute?
- Why did the AI Safety Institute become the AI Security Institute?
- What does this lineage mean for UK research funders and universities?
- Common questions on the UK’s AI governance machinery
What was the UK’s Foundation Model Taskforce?
The Foundation Model Taskforce was announced on 24 April 2023, when DSIT and the Prime Minister’s Office committed an initial £100 million to what the government called “an expert taskforce to help the UK build and adopt the next generation of safe AI.” The taskforce was explicitly modelled on the Vaccine Taskforce’s fast-moving, mission-led structure rather than a standard civil-service directorate.
Tech investor Ian Hogarth was appointed to chair the taskforce on 18 June 2023. Its founding brief covered three strands: building sovereign UK capability in foundation models, procuring pilots for public-service applications, and researching model safety. That third strand — safety research — is the only one that survived into the institution’s later names.
Why did it become the Frontier AI Taskforce, then the AI Safety Institute?
On 7 September 2023, the government renamed the unit the Frontier AI Taskforce, a change confirmed in a written statement to Parliament (HCWS1054, 19 September 2023). The rename narrowed the public framing from “building” foundation models to “evaluating” frontier ones — models at the most capable end of the spectrum.
Two months later, at the AI Safety Summit hosted at Bletchley Park in November 2023, the Frontier AI Taskforce was formally established as a permanent body: the AI Safety Institute. Its stated mission, per its GOV.UK organisation page, was “to minimise surprise to the UK and humanity from rapid and unexpected advances in AI.” Ian Hogarth continued as chair. Capability-building and public-sector procurement — the taskforce’s original first two strands — were absorbed back into DSIT proper, leaving the Institute a narrower, research-and-evaluation-only mandate.
The UK’s move was not isolated. At the AI Seoul Summit in May 2024, ten jurisdictions — the UK, the US, the EU, Japan, Singapore, South Korea, Canada, France, Kenya and Australia — signed the Seoul Statement of Intent, launching the International Network of AI Safety Institutes to align testing methodologies. The network’s first in-person meeting followed in San Francisco on 20–21 November 2024.
Why did the AI Safety Institute become the AI Security Institute?
In February 2025, Technology Secretary Peter Kyle announced at the Munich Security Conference that the AI Safety Institute would be renamed the AI Security Institute (AISI). The change was not cosmetic. The Institute’s remit was redrawn to focus on “serious AI risks with security implications” — chemical, biological and cyber misuse, and AI-enabled crime, including fraud and child sexual abuse material — and a new criminal-misuse research team was created within it.
Notably, the rebranded Institute’s remit explicitly excludes bias and freedom-of-speech concerns, which had sat within the AI Safety Institute’s broader original scope. The US made a parallel move at almost the same moment: its NIST-housed AI Safety Institute became the Center for AI Standards and Innovation (CAISI), reflecting a wider Anglo-American pivot from precautionary “safety” framing toward narrower “security” framing.
| Name | Established | Chair / lead | Core focus |
|---|---|---|---|
| Foundation Model Taskforce | April 2023 | Ian Hogarth (from June 2023) | Capability-building, procurement, safety research |
| Frontier AI Taskforce | September 2023 | Ian Hogarth | Evaluation of frontier-model risk |
| AI Safety Institute | November 2023 (Bletchley Park) | Ian Hogarth | Safety research and model evaluation |
| AI Security Institute (AISI) | February 2025 (Munich) | DSIT-appointed leadership | Cyber, chem-bio, and criminal-misuse risk; excludes bias/free-speech remit |
What does this lineage mean for UK research funders and universities?
For research administration teams, the practical consequence of four names in under two years is that the government counterpart for AI-safety dialogue has changed shape repeatedly, and its funding role has narrowed each time. Three points matter for institutional engagement.
- AISI is not the Alan Turing Institute. The two are frequently conflated in university communications. AISI is a DSIT research unit with a security-and-evaluation mandate; the Turing Institute is the UK’s separately constituted national institute for data science and AI, with a broader research-council relationship. Research offices briefing academic leadership should not treat the two as interchangeable stakeholders.
- Funder co-delivery already exists. EPSRC and Innovate UK — both part of UK Research and Innovation — partnered with the AI Safety Institute to deliver the Systemic Safety Grants programme: an initial £4 million tranche (of £8.5 million earmarked across later phases), funding around 20 projects of up to £200,000 each, with proposals due 26 November 2024 and awards confirmed by February 2025. This is the clearest precedent for how AISI-adjacent funding reaches universities: through UKRI council co-delivery, not direct AISI grant lines.
- AISI holds no statutory or regulatory power. Unlike the Information Commissioner’s Office or a licensing regulator, AISI cannot compel model access, issue penalties, or set binding standards — its leverage is voluntary pre-deployment testing agreements with frontier labs. Universities citing AISI in governance or ethics documentation should describe it accurately as an evaluation and research body, not a regulator.
The direction of travel — from open-ended “safety” to bounded “security” — also signals where new funding calls are likely to concentrate: cyber-misuse and biosecurity-adjacent AI research, rather than the bias, fairness and societal-impact work that sat inside the Institute’s original remit. Research offices tracking DSIT and UKRI calls should expect that broader societal-AI-risk work will increasingly need to find funders outside the AISI lineage altogether, most likely through EPSRC’s standard responsive-mode routes or ESRC-led programmes.
Common questions on the UK’s AI governance machinery
Did the UK change the name of the AI Security Institute?
Yes. The UK’s AI Safety Institute was renamed the AI Security Institute in February 2025, announced by Technology Secretary Peter Kyle at the Munich Security Conference. The change narrowed the Institute’s remit to security-relevant risks — cyber, chemical, biological and criminal misuse — while dropping bias and free-speech concerns from scope.
Who funds the AI Safety Institute?
AISI is funded through DSIT’s departmental budget, originally drawn from the £100 million committed to the Foundation Model Taskforce in April 2023. Some research activity is co-delivered with UK Research and Innovation councils, such as the EPSRC/Innovate UK-backed Systemic Safety Grants programme, rather than funded solely through AISI’s own budget line.
Who is the head of the AI Safety Institute UK?
Ian Hogarth, a technology investor and entrepreneur, chaired the organisation continuously from June 2023 through its Foundation Model Taskforce, Frontier AI Taskforce and AI Safety Institute phases. Its advisory board has included figures such as Turing Award laureate Yoshua Bengio, reflecting an international scientific-advisory model rather than a purely civil-service structure.
What is the UK AI Security Institute?
The AI Security Institute is a research organisation within DSIT that evaluates the security risks of advanced AI models, rather than a regulator. Per its own mission statement, it exists “to equip governments with a scientific understanding of the risks posed by advanced AI,” working through voluntary testing arrangements with frontier AI developers.
Institutions building AI governance guidance should track AISI’s evolving remit alongside — not instead of — the Alan Turing Institute, UKRI’s research-council funding calls, and any forthcoming statutory AI legislation, since none of these bodies currently holds the full picture on its own. The machinery will very likely change name again; the underlying funding relationships, not the current label, are what research offices should document and monitor.