Every AI Policy Ever · open in the interactive register

Published (non-binding, current)Non-binding · Report (government-commissioned)

International AI Safety Report 2026

Record INT-10 · DSIT 2026/001

Issued by
Independent expert panel chaired by Yoshua Bengio, with a secretariat at the UK AI Security Institute and Mila – Quebec AI Institute; published by the UK Government (DSIT)
Jurisdiction
International AI Safety Report (AI Safety Summit process)
Level · region
International · Global
Published / adopted
2026-02-03
Language
English
Official identifier
DSIT 2026/001

Summary

The second International AI Safety Report, guided by over 100 experts with nominees of over 30 countries and international bodies. It narrows its scope to emerging risks at the frontier of general-purpose AI: malicious use (AI-generated content and crime, manipulation, cyberattacks, biological and chemical weapons), malfunctions (reliability, loss of control) and systemic risks (labour markets, human autonomy). It describes an 'evaluation gap', as pre-deployment tests do not reliably predict real-world risk and models increasingly tell test settings from deployment, and covers defence-in-depth safeguards and Frontier AI Safety Frameworks. It makes no specific policy recommendations.

Safety topics

general-purpose / frontier AIcatastrophic riskevaluation & testingresearch prioritiesinternational cooperation

Status notes

Published on 3 February 2026 as the second edition of the report series created after the 2023 AI Safety Summit at Bletchley Park; the UK minister's foreword says it would be showcased at the India AI Impact Summit. Its scope is narrower than that of the 2025 report (INT-09), which also covered bias, environmental impacts, privacy and copyright, and it is meant to complement the UN's Independent International Scientific Panel on AI. It records that in 2025, 12 companies published or updated Frontier AI Safety Frameworks and that most risk-management initiatives remain voluntary, though a few jurisdictions are beginning to make some practices legal requirements. Translations were published in the other five official UN languages, and a 2026 Report: Extended Summary for Policymakers and an Executive Summary were published on the same day.

Sources

Open the official source https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026

Secondary source: https://internationalaisafetyreport.org/sites/default/files/2026-02/international-ai-safety-report-2026.pdf

Local copy

Archived. Last attempt 2026-10-17 16:29:51 UTC

Official source (third-party copy, not published) archive/INT-10/2026-10-11/official.html
Retrieved 2026-10-11 13:57:32 UTC
SHA-256 ac6df460ac4e8ce14bc0fb6d45ed886c901ca45da93ad8190df6b2a7777e4d81
Linked document (PDF) (third-party copy, not published) archive/INT-10/2026-10-11/document-1.pdf
Retrieved 2026-10-11 13:57:37 UTC
SHA-256 e2a35f439cafade360e4df4a2f91f217a16174bd1231f97e64f7208ffa93e225
Linked document (PDF) (third-party copy, not published) archive/INT-10/2026-10-11/document-2.pdf
Retrieved 2026-10-11 13:57:42 UTC
SHA-256 765570f1e573e28a2cedbef5be4abe3f3edc4fe01c51578f81979dc0998e655b
Secondary source (third-party copy, not published) archive/INT-10/2026-10-11/secondary.pdf
Retrieved 2026-10-11 13:57:44 UTC
SHA-256 50b244ddf39fb9189d44ccd990ba423ebe8927d8ffdce0f5aac5882c74cbc790

Verification

A
The instrument's own official text, or an official register, was stored and confirms the title, date and status. checked 2026-10-17 · official-url-fetched
Stored: the report's publication page on internationalaisafetyreport.org (dated 3 February 2026, with the full text and the citation 'DSIT 2026/001'), the English PDF linked from it (220 pages, 'Published in February 2026 by the UK Government'), the French translation, an earlier English PDF from the same site (the record's secondary URL; its text is the same apart from an extra cover page), and in this run the site's Publications page. They confirm the title, date, chair, secretariat (UK AI Security Institute and Mila), research series number and content, the Extended Summary for Policymakers and the Executive Summary dated 3 February 2026, and that no later edition is listed. The earlier note that Yoshua Bengio co-chairs the UN Independent International Scientific Panel on AI rests on no stored file and was dropped.
Second check: Independent check by Claude Code (claude-sonnet-5-5), 2026-10-17: 8 field(s) supported.

History

  1. 2026-10-18related: [{"slug": "international-ai-safety-report-2025", "id": "INT-09"}, {"slug": "new- → [{"slug": "international-ai-safety-report-2025", "id": "INT-09"}, {"slug": "new- — two-way link added: INT-08 link(s) to INT-102026-10-18-fix-related-links · Claude Code (claude-opus-5-5)
  2. 2026-10-17notes: Published 3 Feb 2026, before the India AI Impact Summit. An Extended Summary for → Published on 3 February 2026 as the second edition of the report series created — rewritten from the stored text: Bletchley mandate, India AI Impact Summit, narrower scope than 2025, complementarity with the UN Scientific Panel, Frontier AI Safety Frameworks, translations; the Extended Summary and Executive Summary from the Publications page stored in this run; 'Bengio also co-chairs the UN Independent International Scientific Panel on AI' and the date of that panel's report removed, since no stored file of this record supports them2026-10-17-recheck-international-2 · Claude Code (claude-opus-5-5)
  3. 2026-10-17summary: The second edition, written by over 100 experts, focuses on emerging risks at th → The second International AI Safety Report, guided by over 100 experts with nomin — checked against the key developments, the executive summary and 'About this Report': 'guided by' over 100 experts (the report's wording), the risk sub-sections named, 'models increasingly detect when they are being tested' replaced by the report's 'distinguish between test settings and real-world deployment', the claim about jurisdictions moved to the notes in the report's own terms2026-10-17-recheck-international-2 · Claude Code (claude-opus-5-5)
  4. 2026-10-17issuing_body: Independent expert panel chaired by Yoshua Bengio; secretariat UK AI Security In → Independent expert panel chaired by Yoshua Bengio, with a secretariat at the UK — same wording as INT-09: the secretariat names (AI Security Institute and Mila) from the contributor list, and the publisher from the colophon 'Published in February 2026 by the UK Government'2026-10-17-recheck-international-2 · Claude Code (claude-opus-5-5)

Related records

International AI Safety Report 2025 International AI Safety Report (AI Safety Summit process) · Superseded

New Delhi Declaration AI Impact Summit (New Delhi, 2026) · Published (non-binding, current)

UN AI Scientific Panel Preliminary Report United Nations · Published (non-binding, current)

International AI Safety Report (interim, 2024) International AI Safety Report (AI Safety Summit process) · Superseded

Citation

Independent expert panel chaired by Yoshua Bengio, with a secretariat at the UK AI Security Institute and Mila – Quebec AI Institute; published by the UK Government (DSIT). International AI Safety Report 2026 [DSIT 2026/001]. 2026-02-03. https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026. Status: Published (non-binding, current); verified 2026-10-17 (level A). Local copy archive/INT-10/2026-10-11/official.html retrieved 2026-10-11 13:57:32 UTC, SHA-256 ac6df460ac4e8ce14bc0fb6d45ed886c901ca45da93ad8190df6b2a7777e4d81. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record INT-10.

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.