The UNESCO Recommendation on AI ethics sets values and policy actions for risk governance without approving any product or provider.
Ethical governance follows the risk of the use
The Recommendation establishes a global ethical policy framework centred on human rights, dignity, environmental flourishing, diversity, fairness, transparency, responsibility and human oversight. It directs policy action by Member States. It does not certify an algorithm, authorize a deployment or displace binding law.[S1]
The Recommendation combines values, principles and policy actions rather than prescribing one technical control set. Human rights, fairness, privacy, transparency, accountability and human oversight become operational questions only when connected to a defined system and use. An AI tool used to schedule rooms presents a different educational and rights profile from a system influencing admission, assessment, progression or learner welfare.[S1]
Values, principles and education policy action
The instrument covers proportionality, safety, privacy, data governance, explainability, accountability, literacy and impact assessment. Its education and research policy area addresses curricula, capacity and responsible research. Institutional review can translate these themes into use-case inventories, assigned authority, impact records, human review and accessible contestability.[S1]
The education and research policy area sits within a broader governance framework that includes ethical impact assessment, data policy, communication, health, environment and economic considerations. A credible institutional record identifies the system version, intended use, affected groups, decision authority, data sources and route for challenge. Testing then addresses performance and unequal effects under the conditions in which the system is actually used.[S1]
Risk evidence changes over the life of a system. Pre-deployment assessment addresses foreseeable effects; monitoring addresses actual operation; incident and appeal records expose failures; and change control determines whether earlier evidence remains applicable. The resulting file can distinguish a risk accepted by an authorized body from a risk that was never identified or was transferred contractually without effective oversight.[S1]
Evidence for consequential education uses
The Recommendation addresses States and the AI lifecycle across sectors. Education relevance varies between teaching support, assessment, admissions, proctoring, analytics, research and administration. Each use needs its own affected population, purpose and legal basis.[S1]
The Recommendation is addressed to UNESCO Member States and is designed for varied legal, cultural and development contexts. It can guide institutional governance without replacing national law. Responsibility may be distributed among the developer, supplier, education provider and public authority. The evidence map needs to show which actor controls model design, data, deployment settings, human review and corrective action.[S1]
From ethical principle to verifiable control
Ethical alignment language is not evidence of performance or remedy. A principles document cannot answer whether data are representative, outputs are accurate or a specific decision is lawful. Risk controls also require testing after deployment.[S1]
An ethics policy, impact-assessment form or statement of human oversight does not prove effective governance. Human review has limited value when the reviewer lacks information, authority or time to change a result. Aggregate performance may also conceal error rates for smaller groups. The Recommendation identifies a governance direction, while use-specific evidence determines whether a particular control works.[S1]
Relationship with ICEQC certification
ICEQC cites this UNESCO recommendation as external policy authority. It remains distinct from ICEQC’s voluntary conformity documents and does not indicate UNESCO endorsement. The historical record contains no subsequently issued ICEQC requirement.
ICEQC may consider the Recommendation as external policy evidence relevant to an AI-related certification scope. It neither attributes UNESCO recognition to ICEQC nor converts every policy action into a certification requirement. The record is bounded by the publication position and does not import later controlled clauses.