OECD Principles on AI: The Global Ethical Foundation

A deep dive into the intergovernmental principles that shaped nearly every major AI framework that followed

What Is It?

The OECD Principles on Artificial Intelligence are a set of non-binding recommendations adopted by the Organisation for Economic Co-operation and Development in May 2019. They were the first intergovernmental standard on AI to be adopted, and they established shared ethical norms intended to guide AI development and governance across member nations and beyond.

The principles were updated in 2023 and 2024 to address AI systems that continue to evolve after deployment and those that incorporate generative AI.

Source: oecd.ai/en/ai-principles

Why Were They Created?

By 2019, AI was advancing rapidly while global governance was nearly nonexistent. Individual countries were beginning to develop their own approaches, but fragmentation was a growing concern — especially for cross-border trade, data sharing, and the global supply chains that AI systems increasingly depend on.

The OECD Principles were created to:

  • Establish a shared vocabulary and definitional baseline for what AI systems are
  • Articulate common ethical expectations that member governments and the private sector could align around
  • Provide a foundation for national AI strategies and regulations
  • Promote trustworthy AI that respects human rights and democratic values
  • Enable policy interoperability across borders

The Five Core Principles

1. Inclusive Growth, Sustainable Development, and Well-Being

AI should benefit people and the planet. It should be developed in ways that promote inclusive growth and sustainable development, and address environmental impacts.

2. Human-Centred Values and Fairness

AI systems should respect the rule of law, human rights, and democratic values. They should be fair, non-discriminatory, and designed to protect privacy and data security.

3. Transparency and Explainability

AI actors should be transparent about their AI systems and provide meaningful information about how they work, especially when they affect people’s lives. This includes being able to explain AI-driven outcomes when appropriate.

4. Robustness, Security, and Safety

AI systems should function reliably and safely throughout their lifecycle. Risks should be continuously assessed and mitigated. AI should not be deployed when risks cannot be adequately addressed.

5. Accountability

AI actors — developers, deployers, operators — should be accountable for ensuring that their AI systems operate in accordance with these principles. This includes mechanisms for redress when harm occurs.

Influence and Adoption

The OECD Principles punch well above their non-binding weight:

  • Adopted by the G20 in 2019 — giving them reach to the world’s largest economies
  • Directly influenced the development of the EU AI Act and the NIST AI Risk Management Framework
  • Used by many governments as the definitional baseline for national AI strategies
  • Referenced by the Council of Europe’s Framework Convention on AI (2024) — the first legally binding international AI treaty

The 2023–2024 Updates

The 2023 and 2024 revisions were significant. They clarified the definition of AI systems to explicitly include:

  • Systems that continue to learn and evolve after initial deployment
  • Systems incorporating generative AI capable of producing text, images, code, and other content

This update was important because the original 2019 definition was written before large language models became dominant, and policy discussions were hampered by definitional confusion about what counted as “AI” under the framework.

Honest Limitations

  • The Principles are non-binding — there are no penalties for governments or companies that ignore them
  • They operate at a high level of abstraction — they provide aspirations and direction, not operational guidance
  • Implementation is entirely dependent on national will — the same principles can be interpreted very differently across jurisdictions
  • The 2024 updates, while helpful, are still catching up with the pace of generative AI development

Key Sources

Previous: ISO/IEC 42001 | Next: UNESCO AI Ethics