Every AI Policy Ever · open in the interactive register

Published (non-binding, current)Non-binding · Guidelines / framework

Recommendation on the Ethics of Artificial Intelligence

Record UNESCO-01

Issued by
UNESCO General Conference (41st session)
Jurisdiction
UNESCO
Level · region
International · Global
Published / adopted
2021-11-23
Language
Arabic, Chinese, English, French, Russian, Spanish

Summary

UNESCO's global standard-setting instrument on the ethics of AI, which Member States are to apply on a voluntary basis. It says that unwanted harms (safety risks) and vulnerabilities to attack (security risks) should be avoided and addressed throughout the life cycle of AI systems, that an AI system can never replace ultimate human responsibility and accountability, and that, as a rule, life and death decisions should not be ceded to AI systems. AI should not be used for social scoring or mass surveillance. Member States are asked to introduce ethical impact assessment and liability frameworks and to develop standards with measurable, testable levels of safety and transparency.

Safety topics

human oversightaccountability & liabilitytransparency & labellingprohibited practicesevaluation & testingpublic-sector use

Status notes

Adopted on 23 Nov 2021 at the 41st session of the General Conference, meeting in Paris from 9 to 24 Nov 2021; UNESCO says its 193 Member States adopted it. The General Conference mandated its preparation in November 2019 by 40 C/Resolution 37. To support implementation, the Recommendation commits UNESCO to develop a readiness assessment methodology and a methodology for ethical impact assessment of AI technologies.

Sources

Open the official source https://www.unesco.org/en/legal-affairs/recommendation-ethics-artificial-intelligence

Secondary source: https://www.unesco.org/en/artificial-intelligence/recommendation-ethics

Local copy

Archived. A linked PDF could not be downloaded (https://unesdoc.unesco.org/ark:/48223/pf0000381133/PDF/381133eng.pdf.multi: HTTP 403 Forbidden). Last attempt 2026-10-17 13:37:43 UTC

Official source (third-party copy, not published) archive/UNESCO-01/2026-10-11/official.html
Retrieved 2026-10-11 13:57:45 UTC
SHA-256 850609fd45e77fbf462813953083a42309dddbd4cc078acd5e6e3b79768e3dc0
Secondary source (third-party copy, not published) archive/UNESCO-01/2026-10-11/secondary.html
Retrieved 2026-10-11 13:57:47 UTC
SHA-256 8b93490f1db4887ce7f102d1bf4ad428c353ee6c5543f3f2bfecfae14ec4efda

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 copies (unesco.org, 11 Oct 2026; the archiver found them unchanged on 17 Oct 2026): UNESCO's legal-affairs page for the instrument, which carries the full text of the Recommendation, and UNESCO's AI ethics page, the record's secondary source. The legal-affairs page confirms the title, the adoption on 23 November 2021 in Paris at the 41st session of the General Conference, the voluntary application by Member States, the six language versions and the provisions in the summary. The PDF on unesdoc.unesco.org could not be stored (robots.txt on 11 Oct 2026, HTTP 403 on 17 Oct 2026); the page's text was used instead. The secondary page states that UNESCO's 193 Member States adopted it. The earlier summary's 'testing before market release of high-risk systems' is not in the text and was removed, as was the topic 'risk-based classification'.
Second check: Independent check by Claude Code (claude-sonnet-5-5), 2026-10-17: 6 field(s) supported.

History

  1. 2026-10-18related: [{"slug": "oecd-recommendation-ai", "id": "OECD-01"}] → [{"slug": "oecd-recommendation-ai", "id": "OECD-01"}, {"id": "AR-01", "slug": "a — two-way link added: AR-01, GH-01, INT-17, JM-01, JO-01, PK-01, RS-01, SN-01, ZM-01 link(s) to UNESCO-012026-10-18-fix-related-links · Claude Code (claude-opus-5-5)
  2. 2026-10-17notes: Adopted 23 Nov 2021 at the 41st session of the General Conference in Paris (9-24 → Adopted on 23 Nov 2021 at the 41st session of the General Conference, meeting in — Reworded to the text: 40 C/Resolution 37 mandate, the 193 Member States from UNESCO's page, and the readiness and ethical impact assessment methodologies as commitments in para. 49.2026-10-17-recheck-international-1 · Claude Code (claude-opus-5-5)
  3. 2026-10-17safety_topics: ["human oversight", "accountability & liability", "transparency & labelling", "p → ["human oversight", "accountability & liability", "transparency & labelling", "p — 'risk-based classification' replaced by 'evaluation & testing': the Recommendation has no risk tiers but asks for ethical impact assessment, auditing and standards with testable levels of safety.2026-10-17-recheck-international-1 · Claude Code (claude-opus-5-5)
  4. 2026-10-17summary: A global standard-setting instrument on AI ethics adopted by all UNESCO Member S → UNESCO's global standard-setting instrument on the ethics of AI, which Member St — The old summary's 'testing before market release of high-risk systems' is not in the Recommendation; rewritten from the stored text (paras. 26, 27, 36, 50, 53, 64, 68 and the preamble's voluntary-basis clause).2026-10-17-recheck-international-1 · Claude Code (claude-opus-5-5)

Related records

OECD AI Principles OECD · Published (non-binding, current)

Argentina Trustworthy AI Recommendations (2023) Argentina · Published (non-binding, current)

Ghana National AI Strategy 2025-2035 Ghana · Published (non-binding, current)

Montevideo Declaration Montevideo ministerial summit on AI (2024) · Published (non-binding, current)

Jamaica National AI Task Force Policy Recommendations Jamaica · Published (non-binding, current)

Jordan National AI Ethics Charter Jordan · Published (non-binding, current)

Pakistan National AI Policy 2025 Pakistan · Published (non-binding, current)

Serbia AI Ethics Guidelines (2023) Serbia · Published (non-binding, current)

Senegal National AI Strategy (SNIA, 2023) Senegal · Published (non-binding, current)

Zambia National AI Strategy Zambia · Published (non-binding, current)

Citation

UNESCO General Conference (41st session). Recommendation on the Ethics of Artificial Intelligence. 2021-11-23. https://www.unesco.org/en/legal-affairs/recommendation-ethics-artificial-intelligence. Status: Published (non-binding, current); verified 2026-10-17 (level A). Local copy archive/UNESCO-01/2026-10-11/official.html retrieved 2026-10-11 13:57:45 UTC, SHA-256 850609fd45e77fbf462813953083a42309dddbd4cc078acd5e6e3b79768e3dc0. Every AI Policy Ever, version 1.18 (snapshot 2026-10-18), record UNESCO-01.

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.