The OECD Recommendation on Artificial Intelligence is a non-binding international AI instrument adopted by the OECD Council on 22 May 2019 as Recommendation of the Council on Artificial Intelligence (OECD/LEGAL/0449) and updated on 3 May 2024. It sets out five value-based principles and five recommendations for national policies. Although it carries no enforcement mechanism, it has served as a reference point for subsequent international AI frameworks, including the Bletchley Declaration, the G7 Hiroshima AI Process, the G20 AI Principles, and many national strategies. Per G'sell (2024): "Most of these international initiatives have been strongly influenced by the work of the OECD's 'Recommendation on Artificial Intelligence.'"
Scope and principles
Section 1 of the Recommendation sets out five value-based principles for responsible stewardship of trustworthy AI:
- Inclusive growth, sustainable development, and well-being — AI should benefit people and the planet.
- Human rights and democratic values, including fairness and privacy — respect rule of law, human rights, diversity, fairness, social justice; avoid discriminatory bias.
- Transparency and explainability — disclose that AI is in use; provide meaningful information; enable affected persons to understand and challenge outcomes.
- Robustness, security, and safety — function appropriately across the lifecycle; risk management for bias, cybersecurity, and harm.
- Accountability — AI actors (developers, deployers, operators) accountable for proper AI system functioning.
Section 2 sets out five recommendations directed at national policies:
- Invest in AI research and development.
- Foster a digital ecosystem for AI — data infrastructure, compute, open standards.
- Shape an enabling policy environment — regulatory sandboxes, life-cycle-appropriate frameworks.
- Build human capacity — skills, inclusive labor-market preparation.
- International cooperation — share data, support multi-stakeholder initiatives, shared metrics.
2024 update
The May 2024 update, adopted at the OECD Council meeting in Paris, refined the instrument's definitions and incorporated generative AI developments. The changes included explicit recognition of foundation models and general-purpose AI systems, expanding the scope beyond the narrow and specific-purpose AI assumptions of the 2019 text; enhanced language on misinformation, disinformation, and deepfakes; strengthened privacy and intellectual property provisions; emphasis on transparency of training data and capabilities; and more detailed risk-management recommendations covering safety, security, and sustainability. The interval between the 2019 adoption and the 2024 update was five years, over which frontier AI capabilities changed substantially.
Adherents
As of 2024, all 38 OECD members adhere to the Recommendation. Non-OECD adherents include Argentina, Brazil, Egypt, Malta, Peru, Romania, Singapore, Ukraine, and Uruguay, among others, giving the instrument a footprint broader than OECD membership alone.
Downstream influence
The Recommendation has been used as a template for several subsequent instruments. The G20 AI Principles (2019) were largely based on its principles. The Bletchley Declaration (November 2023), the G7 Hiroshima Code of Conduct (October 2023), and the Seoul Frontier AI Safety Commitments (May 2024) each build on its framing. National strategies in Japan, South Korea, Brazil, Canada, Singapore, Saudi Arabia, and other countries explicitly reference or draw on it. Recurring elements of AI governance vocabulary — transparency, human oversight, non-discrimination, and accountability — were consolidated into a shared formulation by this document.
Implementation across adherents varies in form and degree. The EU AI Act operationalizes the principles in binding detail; the US NIST AI RMF operationalizes them voluntarily; some adherents have done little. The Recommendation itself remains principle-level: it does not specify compute thresholds, capability tiers, or prohibited practices, and its adherence is political rather than legal.
OECD AI Policy Observatory (OECD.AI)
The OECD maintains the OECD.AI Policy Observatory (oecd.ai), a live data hub covering national AI strategies, AI-related policies, and AI-incident tracking. The Observatory is used as a cross-jurisdictional reference source for national AI strategies and is cited throughout G'sell's report.
Relationships
- instance-of: Regulatory Typology: Self-Regulation, Co-Regulation, Traditional Government Regulation (international-level soft-law).
- supports: Regulating Under Uncertainty: Governance Options for Generative AI (G'sell, 2024) (cited as the intellectual ancestor of most international initiatives); national AI strategies across all adherents.
- related: Council of Europe Framework Convention on AI and Human Rights, Democracy, and the Rule of Law (CoE's binding counterpart to the OECD's soft-law framework); The Bletchley Declaration (AI Safety Summit, 1–2 November 2023); G7 Hiroshima Code of Conduct for Advanced AI (2023); Frontier AI Safety Commitments (Seoul, 2024); Paris AI Action Summit Declaration (2025); OECD — Assessing Potential Future AI Risks, Benefits, and Policy Imperatives (later OECD work on AI risks and benefits).
- depends-on: Regulating Under Uncertainty (the epistemic premise for principle-based soft-law).
See also
- OECD — Assessing Potential Future AI Risks, Benefits, and Policy Imperatives — the OECD's 2024 policy imperatives report, which builds on and extends the Recommendation.
- Regulating Under Uncertainty: Governance Options for Generative AI (G'sell, 2024) — Ch. 6 covers the OECD's role in the international initiatives ecosystem.