What they are
The OECD AI Principles are set out in the OECD Council Recommendation on Artificial Intelligence (OECD/LEGAL/0449), adopted on 22 May 2019. They are the first intergovernmental standard on AI and are non-binding. The G20 AI Principles are drawn from them.
Five values-based principles
- Inclusive growth, sustainable development and well-being
- Human rights and democratic values, including fairness and privacy
- Transparency and explainability
- Robustness, security and safety
- Accountability
Five recommendations for governments: invest in AI R&D; foster an inclusive AI-enabling ecosystem; build an interoperable governance and policy environment; build human capacity and prepare for labor transition; and cooperate internationally.
Updates
| Date | Change |
|---|---|
| 8 Nov 2023 | Revised definition of “AI system” (covering generative AI) |
| 3 May 2024 | Ministerial update: misinformation and disinformation, misuse, safe override and decommissioning, clearer transparency, responsible business conduct, interoperable governance, environmental sustainability |
The OECD lists 47 adherents: the 38 OECD members, the EU, and eight non-members (Argentina, Brazil, Egypt, Malta, Peru, Romania, Singapore, Ukraine). We found no 2025-2026 revision.
The OECD AI system definition is widely referenced, including in the EU AI Act.
Reuse terms: why “check required”
We could not read the OECD terms and conditions page itself. Japan’s National Diet Library reports that from 1 July 2024 OECD publications are, in principle, released under CC BY 4.0 as part of OECD open access, with some restrictions for older works (for example, commercial translations need permission). Treat this as a credible secondary report, and check the licence notice on each document: the legal instrument, explanatory pages, and OECD.AI datasets may each carry different terms.
Using the principles in AI and RAG
- Index the legal instrument (OECD/LEGAL/0449) with its amendment dates.
- Separate principles, commentary, and data and record each one’s licence.
- Map your AI policy to the five principles for gap checks.
- Attribute the OECD and note edits.
- Label as non-binding.
Common mistakes
- Using the 2019 text without the 2023 and 2024 amendments.
- Assuming all OECD.AI content shares one licence.
- Treating the principles as legal obligations.
- Citing an old adherent count.
Checklist
- Amended text indexed with dates
- Licence recorded per document type
- Policy mapped to five principles
- Attribution in outputs
- Non-binding label
How AskDona helps
AskDona maps your AI policies to the OECD principles and shows evidence for each, with citations to your documents and the instrument.
Run a sample OECD principles alignment check
AskDona shows the evidence; your governance team decides.
Editor’s note. This article is general information, not legal advice. OECD licensing relies on a secondary source. The “check required” rating is our first-pass judgment.
FAQ
When were the OECD AI Principles adopted? 22 May 2019; updated in November 2023 and May 2024.
How many countries adhere? The OECD lists 47 adherents.
Are they binding? No.
Can I reuse them? OECD publications are reported to be CC BY 4.0 since July 2024; check each document’s notice.
References
All pages opened and checked by October 26, 2026.
| Source | What it supports |
|---|---|
| OECD.AI: AI Principles | Principles, 47 adherents, dates |
| OECD: Legal instrument OECD/LEGAL/0449 (PDF) | Amendment dates, 2024 changes |
| National Diet Library: OECD open access report | CC BY 4.0 from July 2024 |