What the OECD AI Principles Ask of Governments and Firms

The OECD AI Principles are, formally, the Recommendation of the Council on Artificial Intelligence, adopted on 22 May 2019 and revised on 3 May 2024. It sets five principles for trustworthy AI, addressed to everyone who builds or runs AI systems, and five recommendations for governments, none of them binding.

They matter because of who signed up and who copied them. Fifty-one adherents stand behind them today: all 38 OECD members, 12 other countries and the European Union. In June 2019 the G20 based its own AI Principles on them. Among the nine jurisdictions in the Regulation Tracker, the EU, Japan, South Korea, Australia, the United States, Singapore and Saudi Arabia adhere. The UAE and China do not.

What do the five principles say?

All five speak to "AI actors", meaning anyone with a role in an AI system's lifecycle. Here are the headings as revised in 2024:

1. Inclusive growth, sustainable development and well-being. AI should benefit people and the planet, and since 2024 the text names environmental sustainability.
2. Respect for the rule of law, human rights and democratic values, including fairness and privacy. This now includes "addressing misinformation and disinformation amplified by AI".
3. Transparency and explainability. People should know when they are dealing with an AI system, get plain information about how an output came about where feasible, and be able to challenge an output that harms them.
4. Robustness, security and safety. Systems that risk undue harm should be able to be "overridden, repaired, and/or decommissioned safely as needed".
5. Accountability. AI actors answer for the proper functioning of their systems, keep traceability across datasets, processes and decisions, and run "a systematic risk management approach" across the lifecycle.

What do the five recommendations ask of governments?

The second half of the text asks governments to do five things: (1) invest in AI research and development, including open-source tools and open datasets, (2) foster an inclusive AI-enabling ecosystem, (3) shape an "enabling interoperable governance and policy environment", (4) build human capacity and prepare for changes in the labour market, and (5) co-operate internationally, including on technical standards and comparable indicators. These are policy commitments, not duties on companies.

What changed in the 2024 revision?

Two amendments. On 8 November 2023 the OECD updated its definition. On 3 May 2024 ministers revised the text to deal with generative AI. In the OECD's own summary, the changes cover misinformation and the integrity of information, uses outside a system's intended purpose and misuse, clearer disclosure duties, the ability to override or decommission a system safely, responsible business conduct, interoperability and environmental sustainability. Traceability and risk management moved under the accountability principle. "Inappropriate bias" became "harmful bias", and intellectual property rights joined the list of risks.

The definition now reads: "An AI system is a machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment."

How do the principles reach binding law?

Indirectly, through texts that borrow from them. The clearest case is the EU. Article 3(1) of the AI Act defines an AI system in words that closely follow the OECD definition, and recital 12 says the definition should be "closely aligned with the work of international organisations working on AI". Nowhere does the Act name the OECD, so the link rests on the OECD's own account, and the wording bears it out. What the Act then requires is in a separate guide.

The G20's Tsukuba statement of June 2019 adopted AI principles "drawn from the OECD Recommendation on AI", which it called non-binding. And on 7 February 2025 the OECD launched the reporting framework for the G7's Hiroshima code of conduct, under which developers of advanced AI systems report voluntarily on their practices. A binding international text does exist: the Council of Europe's Framework Convention on AI, a treaty that opened for signature on 5 September 2024. UNESCO's Recommendation on the Ethics of AI, adopted on 23 November 2021, is the other global soft-law text, covered in the UNESCO piece.

The OECD also built tools to track the Principles in practice. Its AI Policy Observatory, launched in February 2020, keeps a living list of government AI policies, and on 5 October 2026 it held 2,553 policy initiatives from more than 80 jurisdictions and organisations, a count worth citing with its date. Hiroshima reports and the text of the Principles sit on the same site, so it is the quickest place to check what a given government has signed up to.

What should a company do with them?

Treat them as the common language of a governance programme. When the EU, South Korea or Colorado ask for transparency, human oversight, traceability and risk management, they ask for things the five principles already name, so a programme built under the five headings needs little translation to meet most binding rules. Accountability is the heading to start with: a named owner for each system, records that let you reconstruct a decision, and a risk process that runs for the life of the system. For the dates and duties that bind in each market, see the Regulation Tracker.

Sources and corrections

• OECD, Recommendation of the Council on Artificial Intelligence, OECD/LEGAL/0449: adoption 22 May 2019, amendments 8 November 2023 and 3 May 2024, adherents list. Read 5 October 2026.
• OECD, press release on the 2024 revision, 3 May 2024; OECD.AI, the AI Principles.
• G20 Ministerial Statement on Trade and Digital Economy, Tsukuba, 8 and 9 June 2019, with the G20 AI Principles annex.
• Regulation (EU) 2024/1689, the AI Act: Article 3(1) and recital 12, Official Journal text.
• OECD, press release on the Hiroshima reporting framework, 7 February 2025.
• OECD.AI policy navigator, count read 5 October 2026.
• Council of Europe, Framework Convention on Artificial Intelligence; UNESCO, Recommendation on the Ethics of Artificial Intelligence.

Changes on 5 October 2026. Rewritten from scratch as a short explainer. The earlier version said 47 countries adhere, but the OECD's record now lists 51 adherents (its own overview page still shows the older count). It also claimed the AI Act codified the OECD principles, yet the Act does not name the OECD, and only its AI system definition follows the OECD wording. An illustrative client story, a slide of influence scores and a maturity estimate without a source were removed. Corrections: hello@askajay.ai.