Recomendación de Ética de la IA de la UNESCO

Resumen ejecutivo

Why the UNESCO Recommendation matters differently from the OECD Principles

The UNESCO Recommendation on the Ethics of AI is often described in the same breath as the OECD AI Principles — both are non-binding international instruments on AI, both predate most national legislation, both articulate values that national frameworks then operationalise. The framing is convenient and structurally misleading.

The two instruments cover different territory and were built for different audiences. The OECD Principles are the conceptual baseline for the 47 jurisdictions that adhere to them, predominantly developed economies plus the European Union and selected partners. The UNESCO Recommendation is the conceptual baseline for the 193 UNESCO Member States, a group that includes essentially every country with a UN seat. The geographic breadth is not a footnote — it is the structural fact that determines what the instrument is for.

The dominant narrative reads UNESCO as the ethics-focused sibling of the OECD: more values-driven, more aspirational, less operational. That framing misses what UNESCO actually built. The Recommendation includes operational chapters on monitoring and evaluation and on means of implementation — chapters that the OECD Principles and most other international instruments do not include. It mandates two specific operational tools: the Readiness Assessment Methodology (RAM), which evaluates a country’s capacity to implement the Recommendation across legal, social, economic, scientific, and technical dimensions, and the Ethical Impact Assessment (EIA), which provides a structured method for assessing AI systems against the Recommendation’s principles. By October 2025, more than 70 countries had engaged with the RAM and over 30 had completed it. That is not aspirational. That is concrete national policy infrastructure being built.

The distinction that matters is this: the OECD Principles work because adherents already have substantial regulatory capacity and AI investment. They provide alignment among jurisdictions that can implement principles through their own well-developed institutional machinery. The UNESCO Recommendation works because adherents include many countries that do not yet have that capacity. It provides the framework within which national capacity can be built — readiness assessment, capacity building tools, operational impact assessment methods. UNESCO is not an alternative to the OECD; it is the implementation infrastructure for the part of the world that the OECD framework does not reach.

This distinction has commercial implications. Organisations operating only in OECD jurisdictions can rely on the OECD vocabulary as the conceptual baseline that flows into binding regional and national instruments. Organisations operating in markets that include non-OECD countries — Latin America beyond the OECD members, much of Africa, parts of Asia, parts of the Middle East — find that the UNESCO Recommendation is the framework national regulators are building from. Multinational AI deployments that cross those geographies need to understand both instruments and where each one anchors the national debate.

—

Subjective and material scope

Who it addresses. The Recommendation addresses two audiences in parallel.

The values, principles, and Policy Action Areas address AI actors broadly — the term UNESCO uses to cover anyone playing a role in the AI system lifecycle: developers, deployers, operators, users, and the institutions and businesses that house them. In this respect the audience overlaps with the OECD framework.

The operational chapters and means of implementation address Member States and their national institutions. These chapters specify what governments are expected to do to give effect to the Recommendation: establish policy frameworks, conduct readiness assessment, deploy ethical impact assessment in procurement and oversight, build capacity, and report on implementation through the UNESCO monitoring cycle.

The inclusion of operational chapters distinguishes the Recommendation from most other UNESCO standard-setting instruments. UNESCO normally produces declarations and recommendations that articulate principles. The Ethics of AI Recommendation went further. It articulated principles and built the operational tools that allow Member States to act on them — RAM, EIA, capacity-building programmes, regional implementation networks, the Global Observatory on Ethics of AI partnered with the Alan Turing Institute, and ongoing peer learning through the Global Forum on the Ethics of AI.

What it covers. Four core values, ten core principles, and eleven Policy Action Areas that together cover the AI lifecycle and its societal context.

The four core values are: respect, protection and promotion of human rights and fundamental freedoms and human dignity; environment and ecosystem flourishing; ensuring diversity and inclusiveness; and living in peaceful, just and interconnected societies.

The ten core principles operationalise the values: 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 and collaboration.

The eleven Policy Action Areas translate values and principles into specific policy domains where governments are expected to act: 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.

The coverage is broader than the OECD framework on three dimensions in particular. Gender is treated as a standalone Policy Action Area, with specific obligations regarding gender bias in AI systems and gender participation in AI development. Culture and language are treated explicitly, addressing the risk that AI systems trained predominantly on Anglophone Western data fail to serve culturally and linguistically diverse populations. Health and social well-being receive their own Policy Action Area, addressing AI in healthcare, mental health, and social welfare contexts.

—

The four core values

Respect, protection and promotion of human rights and fundamental freedoms and human dignity. AI systems must respect, protect, and promote human rights and fundamental freedoms. Human dignity must be a foundational consideration in AI development and deployment.

Environment and ecosystem flourishing. AI systems should contribute to environmental sustainability and ecosystem flourishing. This value precedes the OECD’s 2024 update on environmental sustainability by three years.

Ensuring diversity and inclusiveness. AI must respect, protect, and promote diversity and inclusiveness across cultures, languages, gender, and abilities. AI must not amplify existing inequalities or create new ones.

Living in peaceful, just and interconnected societies. AI must support peaceful and just societies, social cohesion, and the avoidance of conflict-amplifying applications.

—

The ten core principles

Proportionality and do no harm. AI systems should not be deployed beyond what is necessary for legitimate aims, and should not cause harm.

Safety and security. AI systems should avoid unintended harm and prevent misuse.

Fairness and non-discrimination. AI systems should be fair and non-discriminatory, with particular attention to historically marginalised groups.

Sustainability. AI systems should be assessed against environmental, social, and economic sustainability dimensions.

Right to privacy and data protection. AI must respect privacy and data protection through the lifecycle.

Human oversight and determination. Humans should remain ultimately responsible for AI-mediated decisions, with mechanisms for human override.

Transparency and explainability. AI actors should commit to transparency and explainability appropriate to context and stakes.

Responsibility and accountability. AI actors should be responsible and accountable for AI system behaviour and outcomes.

Awareness and literacy. Public awareness and AI literacy should be promoted across the full population, not only among technical communities.

Multi-stakeholder and adaptive governance and collaboration. Governance of AI should engage multiple stakeholders and adapt as the technology evolves.

—

The eleven Policy Action Areas

The Policy Action Areas convert principles into specific domains of national policy action. Each area identifies what Member States are expected to address through legislation, regulation, programmes, or institutional capacity.

Ethical impact assessment. Member States should introduce frameworks for impact assessment of AI systems, identifying benefits, concerns, risks, and risk prevention measures.

Ethical governance and stewardship. Member States should ensure AI systems serve the public interest, with appropriate institutional structures.

Data policy. Comprehensive data governance frameworks aligned with human rights and AI ethics.

Development and international cooperation. International cooperation in AI development, particularly to support countries with less established AI capacity.

Environment and ecosystems. AI policy should account for environmental impact and contribute to ecosystem sustainability.

Gender. Specific attention to gender bias in AI systems and gender participation in AI development.

Culture. Protection and promotion of cultural and linguistic diversity in AI development and deployment.

Education and research. Investment in AI education and research, with particular attention to ethics and governance dimensions.

Communication and information. AI applications in media and information, addressing misinformation, disinformation, and information integrity.

Economy and labour. AI impact on labour markets, with policies for worker protection, transition support, and economic inclusion.

Health and social well-being. AI in healthcare, mental health, and social welfare contexts, with specific safeguards for vulnerable populations.

—

The operational tools — what makes UNESCO different

The Recommendation’s distinctive feature is its commitment to operational implementation. Two tools deserve specific treatment.

Readiness Assessment Methodology (RAM)

The RAM is a macro-level diagnostic tool that helps Member States understand their preparedness to implement the Recommendation. It evaluates national readiness across five dimensions: legal and regulatory, social and cultural, economic, scientific and educational, and technical and infrastructural.

Each dimension uses a combination of qualitative and quantitative indicators. The output is a country report and a tailored roadmap that identifies gaps and prioritises capacity-building actions. The roadmap is co-developed by the Member State, national experts, and UNESCO’s Secretariat. By October 2025, over 70 countries had engaged with the RAM and more than 30 had completed it, including Indonesia (the first Southeast Asian country to complete), the Philippines, Ecuador, Thailand, Malaysia, and others across Latin America, Africa, and Asia.

For multinational organisations operating in adopting countries, the RAM is operationally significant because it shapes the priorities of the national AI strategy that follows. A country that completes the RAM uses the resulting roadmap to define its AI policy agenda for the following years. Understanding where a country sits in its RAM cycle informs how that country’s regulatory environment is likely to evolve.

Ethical Impact Assessment (EIA)

The EIA is a versatile assessment tool applicable in both public and private contexts. In the public sphere, government entities use the EIA to evaluate AI systems being procured against the ethical standards of the Recommendation. In the private sphere, organisations use the EIA as an internal governance tool to assess specific AI deployments against ethical principles.

The EIA structure mirrors the Recommendation: it asks structured questions across the values, principles, and Policy Action Areas, and produces a documented assessment that supports decision-making about AI deployment. UNESCO has integrated EIA training into its capacity-building programmes for civil servants in adopting countries.

The distinction between RAM and EIA matters: RAM is national-scale, used by governments to evaluate their own readiness. EIA is system-scale, used by any actor to evaluate a specific AI system. They operate at different levels of the implementation hierarchy.

—

Governance and implementation infrastructure

The Recommendation is governed through UNESCO’s standard machinery — the General Conference, the Executive Board, and the Secretariat — supplemented by AI-specific structures that have grown around implementation.

Global Forum on the Ethics of AI. The peer learning and policy exchange forum hosted annually. The first edition was hosted by Czechia in February 2024 in the framework of the Czech Presidency of the Council of the European Union. Subsequent editions have rotated to other adopting countries. The Forum is the principal venue where Member States exchange implementation experience.

Global Observatory on Ethics of AI. A digital platform partnered with the Alan Turing Institute that hosts analytical work, country reports, and expert contributions. It serves as the institutional memory of implementation work across the 193 Member States.

AI Ethics Experts Without Borders. A network of experts that supports Member States with technical expertise during RAM and EIA implementation, particularly in countries with limited domestic capacity.

Women 4 Ethical AI. A network specifically focused on gender participation in AI development and gender impact in AI systems, reflecting the Recommendation’s elevation of gender to a standalone Policy Action Area.

Capacity-building programmes. UNESCO has developed AI literacy training for civil servants, deployed through a train-the-trainer model implemented in partnership with the Center for AI and Digital Policy (CAIDP). Training covers AI fundamentals, governance frameworks, RAM, EIA, and the practical considerations for AI in public services including procurement, data governance, and emerging technologies such as agentic systems.

Regional implementation partnerships. UNESCO has established regional partnerships including with the Development Bank of Latin America (CAF) for Latin America and the Caribbean implementation, with European Union support for Southeast Asian implementation, and with various national and regional partners across Africa.

—

Intersections with other instruments

The Recommendation does not operate in isolation. Five intersections shape its operational role:

OECD AI Principles. Complementary, not redundant. The two instruments share substantial conceptual ground — values around human-centred AI, transparency, accountability, safety. The differences are scope (193 UNESCO members vs 47 OECD adherents), operational tooling (UNESCO has RAM and EIA, OECD does not), and emphasis (UNESCO is more developed in gender, culture, and capacity building; OECD is more developed in economic policy framing). Multinational organisations typically engage with both.

Council of Europe Framework Convention on AI. UNESCO informed the conceptual ground that the Council of Europe Convention then translated into legally binding form for its signatories. The Convention is more focused on human rights, rule of law, and democracy than UNESCO’s broader development orientation, but the values heritage runs from UNESCO through OECD to the Convention.

EU AI Act. The AI Act references UNESCO indirectly through its broader rights-based architecture. UNESCO’s elevation of cultural and linguistic diversity intersects with the AI Act’s requirements on transparency for AI-generated content and on bias and fairness in high-risk systems. EU institutions reference both instruments in policy documents.

G7 and G20 frameworks. The G7 Hiroshima Process and the G20 declarations reference UNESCO alongside the OECD. UNESCO is the global ethics counterweight to the more economically focused OECD framework in international AI policy dialogue.

National frameworks in adopting countries. For countries that have completed the RAM, national AI strategies and regulatory frameworks are typically structured around UNESCO’s values, principles, and Policy Action Areas. This is most visible in Latin America, parts of Asia, and parts of Africa, where the UNESCO framework is the dominant conceptual reference rather than the OECD framework.

—

## ⚖️ How Zertia operates within the UNESCO Recommendation

Built around UNESCO AI Ethics from day one

Accreditations and memberships: 🎖️ ANAB-accredited (US) · 🎖️ UKAS process (UK) · 🎖️ ENAC process (EU) · 🏛️ IAPP member · 🏛️ INCITS member · 🏛️ UKAI member · 📜 EU AI Pact signatory

Zertia is an ANAB-accredited AI management system certification body. The UNESCO Recommendation does not produce direct certification engagements because the instrument is non-binding, but it shapes how we approach multi-jurisdictional engagements, particularly with clients operating across markets where UNESCO is the dominant conceptual reference.

Regulatory frameworks — EU AI Act Conformity Assessment, NIST AI RMF Attestation, ISO/IEC 23894 Risk Assessment, Algorithmic Impact Assessment, Pre-Certification Assessment. For multinational clients with operations in Latin America, Southeast Asia, or other markets where the UNESCO RAM has shaped national regulation, our regulatory frameworks team integrates UNESCO-aligned vocabulary into compliance positioning. The Ethical Impact Assessment methodology under UNESCO is structurally compatible with our Algorithmic Impact Assessment service for cases where clients need UNESCO-aligned documentation.

Certification — ISO/IEC 42001, AIUC-1, ISO/IEC 27001, ISO/IEC 27701, ISO/IEC 22301. ISO/IEC 42001 implementation aligns conceptually with UNESCO values and principles, particularly on human oversight, fairness, transparency, and impact assessment. For clients pursuing ISO/IEC 42001 certification in countries that have completed the UNESCO RAM, we structure documentation to support both the ISO certification and any UNESCO-aligned national reporting requirements.

Audit — AI Management System audits, High-Risk AI System audits, AI Model audits, EU AI Act audits, NIST AI risk audits. Audit engagements in markets where UNESCO frames the regulatory baseline include explicit mapping to UNESCO Policy Action Areas where relevant to the engagement scope.

Training — AI Governance, Data Governance, Privacy Governance through Zertia Academy. Programmes include treatment of UNESCO values, principles, and Policy Action Areas alongside OECD, NIST, and ISO frameworks, with specific attention to the gender, cultural diversity, and capacity-building dimensions that UNESCO emphasises more strongly than the other instruments.

Zertia operates from Boston, Madrid, and London, with ANAB accreditation in the United States and active accreditation processes with UKAS (United Kingdom) and ENAC (Spain/EU). Member of IAPP, INCITS, and UKAI. Signatory to the EU AI Pact.

🎯 Take action

🔍 Diagnose your ethics posture	📊 Operationalise the recommendation
Pre-Certification Assessment →	ISO/IEC 42001 Certification →
Independent diagnosis of AI governance against UNESCO ethics principles and the AI management system that operationalises them.	The accredited management system standard that translates UNESCO ethical principles into auditable governance practice.

Discuss multi-jurisdictional AI governance positioning →

—