---
name: unesco-ai-ethics
version: "2021"
date: 2026-07-29
category: ethics
title: UNESCO Recommendation on the Ethics of Artificial Intelligence (2021) — a restatement
language: en
kind: principles
type: restatement
license: CC-BY-4.0
source_url: https://www.unesco.org/en/artificial-intelligence/recommendation-ethics
accessed: 2026-07-29
notes: "Restatement per IFP-22 Section 2.3. The authoritative text is UNESCO document SHS/BIO/PI/2021/1 (unesdoc ark pf0000381137); the unesdoc repository blocks automated retrieval, so the license notice on the publication could not be verified at publication time. UNESCO's standard open-access license (CC-BY-SA 3.0 IGO) would permit verbatim inclusion — a full-text superseding version is a welcome contribution once the text and its license notice can be verified. Until then, a declaration citing this entry is a declaration of the Recommendation as restated here."
---

# UNESCO Recommendation on the Ethics of Artificial Intelligence (2021) — a restatement

This entry restates, in this registry's own words, what UNESCO's *Recommendation on the Ethics of Artificial Intelligence* holds. The Recommendation was adopted by acclamation by UNESCO's General Conference in November 2021 — the first global standard-setting instrument on the ethics of AI, agreed by the organization's member states. Its authoritative text is published by UNESCO (see the frontmatter). What follows is a description of the instrument, not the instrument; anyone applying the Recommendation in earnest should read the original.

## What the instrument is

The Recommendation addresses governments first — it is an intergovernmental normative instrument, meant to be translated into national law and policy — but it speaks to the whole AI lifecycle and all actors in it: research, design, development, deployment, and use. Its through-line is that AI must be governed within the framework of human rights, human dignity, and ecological sustainability, with ultimate responsibility and oversight always remaining human.

## What it holds

**Four values** anchor the instrument: respect, protection, and promotion of human rights, fundamental freedoms, and human dignity; environment and ecosystem flourishing; ensuring diversity and inclusiveness; and living in peaceful, just, and interconnected societies.

**Ten principles** operationalize the values: proportionality and do-no-harm (AI use must be necessary to a legitimate aim and risk-assessed); safety and security; fairness and non-discrimination; sustainability, assessed against evolving goals including the UN Sustainable Development Goals; right to privacy and data protection throughout the lifecycle; human oversight and determination — AI systems must never displace ultimate human responsibility and accountability; transparency and explainability appropriate to context; responsibility and accountability, with AI systems auditable and traceable; awareness and literacy, so publics can understand and question the systems that affect them; and multi-stakeholder and adaptive governance, respecting international law and including diverse participants.

**Eleven policy action areas** translate the principles into concrete fields for member-state action, including ethical impact assessment; ethical governance and stewardship; data policy; development and international cooperation; environment and ecosystems; gender; culture; education and research; communication and information; economy and labour; and health and social well-being.

For the concerns this registry's declarers most often carry: the Recommendation squarely supports evaluating AI in its social and ecological context rather than in isolation; it insists that judgment, authorship, and accountability remain with human beings; and it treats transparency about AI's involvement as an ethical requirement, not a courtesy.

## Standing of this restatement

This restatement was prepared from UNESCO's presentation of the instrument at the source URL, read as of the accessed date in the frontmatter. If UNESCO revises the Recommendation or objects to this characterization, the correction will appear as a new superseding version of this entry; this version remains published because citations to it remain outstanding (IFP-22, Sections 2.3 and 4).