# The UK retitled its AI Safety Institute and put security risks at the centre of its remit On February 14, 2025, the UK renamed its AI Safety Institute the AI Security Institute, adding a criminal-misuse team and foregrounding national-security research. Canonical: https://rangoon.ai/insights/archive-uk-ai-security-institute-february-2025/ Author: Rangoon Editorial (https://rangoon.ai/insights/editorial/) Published: 2026-10-24 Updated: 2026-10-24 Event date: 2025-02-14 Tags: [UK government](https://rangoon.ai/insights/tags/uk-government/), [AI Security Institute](https://rangoon.ai/insights/tags/ai-security-institute/), [AI safety](https://rangoon.ai/insights/tags/ai-safety/), [AI security](https://rangoon.ai/insights/tags/ai-security/) Source note: Historical reporting: on February 14, 2025, the UK government renamed the AI Safety Institute as the AI Security Institute. The same-day press release and Munich speech described a focus on serious security risks, a new criminal-misuse team with the Home Office, and collaboration with defence and cyber bodies. The change did not make the Institute a market-access regulator, prove any model safe, or establish that an evaluation eliminates risk. Image: Security research evidence path. Original Rangoon explanatory diagram: Source boundary: dated primary release reporting; original editorial analysis. No vendor interface, photograph, measurement result, or endorsement implied. ## Key takeaways - The February event changed the Institute’s name and emphasized security-related research rather than creating a general AI approval system. - The stated focus included dual-use science, cyber misuse, crime and loss-of-control questions, each requiring technical evidence. - Testing, research and cross-government collaboration are distinct from a guarantee about a model or deployment. ## The renaming was a public statement about the institute’s research focus [source 1](https://www.gov.uk/government/news/tackling-ai-security-risks-to-unleash-growth-and-deliver-plan-for-change) [source 2](https://www.gov.uk/government/speeches/remarks-made-by-technology-secretary-peter-kyle-at-the-munich-security-conference) On February 14, 2025, the UK government announced that the AI Safety Institute would become the AI Security Institute. The Department for Science, Innovation and Technology said the revised name reflected a focus on serious risks with security implications, including risks related to chemical and biological misuse, cyber attacks, fraud and child sexual abuse. This was a change in public framing and organisational emphasis. It was not a finding that every AI system posed those risks, nor a conclusion that the Institute had solved them. Peter Kyle’s Munich Security Conference speech provided the same-day explanation in more operational language. He said the Institute would not decide questions of bias or freedom of speech, describing its task instead as scientific research into serious emerging risks. That distinction matters in historical coverage. A government research institute can evaluate capabilities and publish evidence without becoming the final authority for every contested social impact, procurement decision or model release. ## A criminal-misuse team and partner links extended the research network [source 1](https://www.gov.uk/government/news/tackling-ai-security-risks-to-unleash-growth-and-deliver-plan-for-change) [source 2](https://www.gov.uk/government/speeches/remarks-made-by-technology-secretary-peter-kyle-at-the-munich-security-conference) The announcement said a new criminal-misuse team would work jointly with the Home Office on crime and security issues. It also pointed to collaboration with the Defence Science and Technology Laboratory, the National Cyber Security Centre and the wider national-security community. These were organisational routes for research and assessment. They did not mean that a developer had automatically supplied a model for testing, that a particular deployment was under government oversight, or that a public user had access to an evaluation result. The speech highlighted cyber, chemical and biological risks, as well as questions about systems acting with limited human instruction. These are categories of inquiry, not a single benchmark score. An evaluation needs a stated model version, test method, access conditions and interpretation limits before it can support a narrow claim. The February announcement establishes the Institute’s intended direction and partnerships; it does not allow a reader to infer a universal technical threshold or a permanent capability ranking. ## The policy language connected security research to adoption without equating the two [source 1](https://www.gov.uk/government/news/tackling-ai-security-risks-to-unleash-growth-and-deliver-plan-for-change) [source 2](https://www.gov.uk/government/speeches/remarks-made-by-technology-secretary-peter-kyle-at-the-munich-security-conference) The government placed the rename alongside its broader AI growth agenda, arguing that confidence and security work could support adoption. That is an economic and policy rationale, not a measurement that confidence had already risen or that deployment risks were uniformly controlled. Organisations considering AI still needed their own testing, data handling decisions, access controls and incident procedures. National research capacity can improve the evidence available to those decisions without taking them over. The press release also discussed an agreement with Anthropic and the government’s Sovereign AI activity. Those parallel announcements should not be read as a certification of a supplier or a guarantee about any product. A partnership can provide channels for research, information sharing or public-service exploration while the details of an individual system’s deployment remain separate. Good historical reporting keeps the organizational announcement, commercial relationships and claims about real-world assurance in different categories. ## The February record supports a remit change, not a safety verdict [source 1](https://www.gov.uk/government/news/tackling-ai-security-risks-to-unleash-growth-and-deliver-plan-for-change) [source 2](https://www.gov.uk/government/speeches/remarks-made-by-technology-secretary-peter-kyle-at-the-munich-security-conference) The new name made the Institute’s direction easier to state: research on the gravest AI risks with national-security and criminal-misuse consequences. It also showed the government locating that research in a network of technical and security bodies. This can shape the kinds of work an institute funds, publishes or coordinates. It does not make the Institute a release gate for all AI systems, and the speech explicitly described the work as scientific rather than a barrier to market access. This historical account does not treat later Institute reports, budgets, statutory proposals or technical findings as February facts. It also avoids claiming that an evaluation makes a system safe or that a model has been approved for a particular use. The evidence supports a limited conclusion: the UK renamed a government AI research body and clarified a security-centered remit, while the scope, methods and consequences of individual evaluations remained questions for separately dated evidence. ## Sources 1. [GOV.UK: Tackling AI security risks to unleash growth and deliver Plan for Change](https://www.gov.uk/government/news/tackling-ai-security-risks-to-unleash-growth-and-deliver-plan-for-change) 2. [GOV.UK: Remarks made by Technology Secretary Peter Kyle at the Munich Security Conference](https://www.gov.uk/government/speeches/remarks-made-by-technology-secretary-peter-kyle-at-the-munich-security-conference) ## Related reading - [UK’s third-party AI assurance roadmap mapped professionalization, information access and a planned innovation fund](https://rangoon.ai/insights/archive-uk-trusted-third-party-ai-assurance-roadmap-september-2025/) - [UK published an AI cyber-security code with 13 voluntary principles and an implementation guide](https://rangoon.ai/insights/archive-uk-ai-cyber-security-code-january-2025/) - [OECD launched the Hiroshima AI reporting framework, making comparable company disclosures the event—not proof of compliance](https://rangoon.ai/insights/archive-oecd-hiroshima-ai-reporting-framework-february-2025/) ## More information - [Documentation index](https://rangoon.ai/llms.txt) - [AI access and policies](https://rangoon.ai/ai/) - [Source repository](https://github.com/hypler-dev/rangoon)