Introducing the AI Safety Institute

Summary

Founding mandate of the UK AI Safety Institute, formed from the Frontier AI Taskforce as the first state-backed organisation focused on advanced AI safety for the public interest. Its mission is to 'minimise surprise to the UK and humanity from rapid and unexpected advances in AI'. Its three core functions are evaluations of advanced AI systems, foundational AI safety research and information exchange. Early evaluations cover dual-use capabilities, societal impacts, system safety and security, and loss of control. The institute is not a regulator and holds no responsibility for release decisions.

Safety topics

Status notes

First published on GOV.UK on 2 November 2023 and presented to Parliament as Command Paper CP 960 in November 2023; GOV.UK records the update of 17 January 2024 as the addition of an AI Safety Institute page. The institute was an evolution of the Frontier AI Taskforce, announced in April 2023, whose research team became its initial core, with Ian Hogarth continuing as Chair. It was to house the secretariat for the international 'State of the Science' report on advanced AI, whose writing group Yoshua Bengio chairs. On 14 February 2025 the institute became the AI Security Institute, with a focus on serious AI risks with security implications; it now operates as a research organisation within DSIT whose stated mission is to equip governments with a scientific understanding of the risks posed by advanced AI.

Sources

Open the official source https://www.gov.uk/government/publications/ai-safety-institute-overview/introducing-the-ai-safety-institute

Secondary source: https://www.aisi.gov.uk/

Local copy

Verification

History

1. 2026-10-18related: [{"slug": "uk-ai-security-institute-2025", "id": "GB-07"}, {"slug": "uk-emerging → [{"slug": "uk-ai-security-institute-2025", "id": "GB-07"}, {"slug": "uk-emerging — two-way link added: GB-04 link(s) to GB-032026-10-18-fix-related-links · Claude Code (claude-opus-5-5)
2. 2026-10-17notes: Built on the Frontier AI Taskforce and launched around the Bletchley Park AI Saf → First published on GOV.UK on 2 November 2023 and presented to Parliament as Comm — rewritten from the stored files: first publication on 2 November 2023, CP 960, what the 17 January 2024 update was, Taskforce origin, Hogarth as Chair, the State of the Science secretariat, the renaming on 14 February 2025 and the institute's current place and mission; the Bletchley summit dates and 'directorate of DSIT' are not in a stored file ('a research organisation within DSIT' is) and were dropped2026-10-17-recheck-europe-2 · Claude Code (claude-opus-5-5)
3. 2026-10-17summary: Founding mandate of the UK AI Safety Institute, the first state-backed body for → Founding mandate of the UK AI Safety Institute, formed from the Frontier AI Task — extended from the stored paper: Taskforce origin, the four evaluation areas of Box 1, not a regulator and no responsibility for release decisions2026-10-17-recheck-europe-2 · Claude Code (claude-opus-5-5)

Related records

UK AI Security Institute United Kingdom · Published (non-binding, current)

Emerging processes for frontier AI safety United Kingdom · Published (non-binding, current)

UK AI regulation government response United Kingdom · Published (non-binding, current)

Citation

Department for Science, Innovation and Technology (DSIT). Introducing the AI Safety Institute [CP 960]. 2023-11-02. https://www.gov.uk/government/publications/ai-safety-institute-overview/introducing-the-ai-safety-institute. Status: Superseded; verified 2026-10-17 (level A). Local copy archive/GB-03/2026-10-11/official.html retrieved 2026-10-11 14:03:10 UTC, SHA-256 173921c42311a0904feba8f8280908bc078fa44942b1efdca8b5cd047155fdfa. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record GB-03.

Improve this record

Know of an amendment, a status change or an error? Report news or an error on GitHub; the form opens with this record filled in.