UNESCO Recommendation on the Ethics of Artificial Intelligence

Overview

Adopted unanimously by all 194 UNESCO Member States at the 41st Session of the General Conference on 23 November 2021, the UNESCO Recommendation on the Ethics of Artificial Intelligence is the first globally applicable normative instrument on AI ethics developed within the United Nations system. It was produced following a two-year consultative process involving more than 100 countries, thousands of experts, and extensive civil society engagement, and was informed by a prior Report of the Ad Hoc Expert Group (AHEG). The Recommendation is structured around four core values-respect, protection and promotion of human rights and fundamental freedoms; environment and ecosystem flourishing; ensuring diversity and inclusiveness; and living together in peace and justice-from which ten principles are derived: proportionality and do no harm; safety and security; fairness and non-discrimination; sustainability; right to privacy and data protection; human oversight and determination; transparency and explainability; responsibility and accountability; awareness and literacy; and multi-stakeholder and adaptive governance. The instrument then translates these principles into 11 thematic policy action areas: ethical impact assessment; ethics governance; data policy; development and international cooperation; environment and ecosystems; gender; culture; education and research; communication and information; economy and labour; and health and social welfare. For each area, specific actions are recommended for Member States and, by extension, for the private sector entities operating within their jurisdictions. A critical operational mechanism introduced by the Recommendation is the Readiness Assessment Methodology (RAM), a self-assessment tool developed by UNESCO to help Member States and organizations evaluate their existing legal, technical, social, and ethical infrastructure against the Recommendation's standards. The RAM has been piloted in multiple countries across Africa, Latin America, and Asia-Pacific and is available to private sector organizations as a voluntary benchmarking instrument. While not legally binding, the Recommendation carries significant normative weight as a consensus instrument of all UN member states and is increasingly cited by national regulators, procurement bodies, and international standard-setting organizations as an authoritative baseline for AI ethics policy. Its explicit emphasis on environmental sustainability, gender equality, cultural diversity, and development equity distinguishes it from OECD-aligned frameworks and broadens its relevance to emerging market enterprises and multinational operations in the Global South.

Key Requirements

• •Member States are called upon to implement ethical impact assessments for AI systems before and during deployment, with particular attention to vulnerable groups.
• •Organizations must ensure meaningful human oversight over AI systems and preserve the right of humans to make final decisions in contexts affecting rights or welfare.
• •Data governance frameworks must embed privacy by design and by default, restrict unnecessary data collection, and ensure data quality and representativeness.
• •Transparency and explainability must be proportionate to the context and impact of the AI system, with particular obligations where automated decisions affect individuals.
• •AI systems must not be used to undermine democratic institutions, freedom of expression, or access to information.
• •Environmental impact of AI systems—including energy consumption and resource use—must be assessed and minimized as part of lifecycle governance.
• •Gender equity and non-discrimination must be embedded in AI design, training data, and deployment practices.
• •All actors in the AI lifecycle bear responsibility and accountability proportionate to their role and influence.
• •UNESCO encourages Member States to adopt or strengthen national AI ethics governance bodies and multi-stakeholder oversight mechanisms.
• •International cooperation, capacity building, and technology transfer are explicitly recommended to address inequities in AI development between high-income and lower-income countries.

Effective Date

2021-11-23