International Scientific Report on the Safety of Advanced AI: interim report

Summary

Interim edition of the first International Scientific Report on the Safety of Advanced AI, written by 75 AI experts with an Expert Advisory Panel nominated by 30 countries, the EU and the UN. It synthesises the evidence on the capabilities of general-purpose AI and sorts its risks into malicious use (fake content, disinformation, cyber offence, biological weapons), malfunctions (bias, loss of control) and systemic risks (labour markets, the AI divide, market concentration, privacy, copyright). It finds that technical methods such as benchmarking, red-teaming and auditing training data can help mitigate risks but all have limitations. It does not discuss policy interventions.

Safety topics

Status notes

Commissioned at the AI Safety Summit at Bletchley Park in November 2023 and chaired by Yoshua Bengio, with a UK Government secretariat hosted by the AI Safety Institute; published in May 2024 ahead of the AI Seoul Summit 2024. Executive summaries were published in Arabic, Chinese, French, Russian and Spanish. The interim report said the final version would be published ahead of the AI Action Summit in France; that full report, the International AI Safety Report 2025 (INT-09), followed in January 2025. On 22 October 2025 the GOV.UK page was updated only to add a link to the report's website, internationalaisafetyreport.org.

Sources

Open the official source https://www.gov.uk/government/publications/international-scientific-report-on-the-safety-of-advanced-ai

Local copy

Verification

History

1. 2026-10-17notes: Published for the AI Seoul Summit. Superseded by the full International AI Safet → Commissioned at the AI Safety Summit at Bletchley Park in November 2023 and chai — rewritten from the stored files: commissioning, secretariat, executive-summary languages, the planned final version and the 2025 report; publication ahead of the AI Seoul Summit 2024 from the website's About page, stored in this run2026-10-17-recheck-international-2 · Claude Code (claude-opus-5-5)
2. 2026-10-17summary: A science-based overview of general-purpose AI capabilities and risks. It covers → Interim edition of the first International Scientific Report on the Safety of Ad — checked against the executive summary: the 75 experts and the 30-country panel added, the three risk categories filled in from the executive summary, 'commissioned under the Bletchley Declaration' replaced (the text says the work was 'commissioned at the Bletchley Park AI Safety Summit'), the absence of policy discussion added2026-10-17-recheck-international-2 · Claude Code (claude-opus-5-5)
3. 2026-10-17official_identifier: – → DSIT 2024/009 — the GOV.UK page gives 'Ref: DSIT research paper series number 2024/009' and the PDF 'Research series number: DSIT 2024/009'; the 2025 report (INT-09) carries its DSIT number in the same field2026-10-17-recheck-international-2 · Claude Code (claude-opus-5-5)
4. 2026-10-17date_last_updated: 2025-10-22 → – — the 22 October 2025 GOV.UK update only added a link to the report's website ('Updated with a link to the external website of the International AI Safety Report'); the report was not amended or reissued, so the field is cleared and the update goes in the notes2026-10-17-recheck-international-2 · Claude Code (claude-opus-5-5)

Related records

Bletchley Declaration AI Safety Summit (Bletchley, 2023) · Published (non-binding, current)

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

International AI Safety Report 2026 International AI Safety Report (AI Safety Summit process) · Published (non-binding, current)

Citation

Independent expert group chaired by Yoshua Bengio; published by UK DSIT, the AI Safety (now Security) Institute and the Cabinet Office. International Scientific Report on the Safety of Advanced AI: interim report [DSIT 2024/009]. 2024-05-17. https://www.gov.uk/government/publications/international-scientific-report-on-the-safety-of-advanced-ai. Status: Superseded; verified 2026-10-17 (level A). Local copy archive/INT-08/2026-10-11/official.html retrieved 2026-10-11 13:57:23 UTC, SHA-256 017f482de89ea041b01ddd42ab3d2eba24b216a1365d988c51136d0508f70237. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record INT-08.

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.