OECD AI Principles
Intergovernmental principles for trustworthy AI adopted by OECD members and partner countries, foundational to subsequent AI governance frameworks.
The OECD AI Principles are the first intergovernmental standard for trustworthy AI, adopted as Recommendation of the OECD Council on Artificial Intelligence on May 22, 2019 by 42 OECD member and partner countries (subsequently expanded). The principles emerged from the OECD AI Group of Experts (AIGO) work commenced 2018 and were endorsed by the G20 at Osaka June 2019, becoming the basis for the G20 AI Principles. The Recommendation establishes five values-based principles for AI actors (inclusive growth, sustainable development and well-being; human-centered values and fairness; transparency and explainability; robustness, security and safety; accountability) and five recommendations for governments (investing in AI R&D; fostering a digital ecosystem for AI; shaping an enabling policy environment; building human capacity and preparing for labor-market transformation; international cooperation). The Principles were updated May 3, 2024 to address generative AI, incorporating considerations including content provenance, misinformation, and intellectual-property concerns. Soft-law instrument: not legally binding but politically authoritative and influential as reference framework for subsequent AI governance instruments globally.
Core components
- Five values-based principles for AI actors: (1) inclusive growth, sustainable development and well-being
- (2) human-centered values and fairness (rule of law, human rights, democratic values, diversity, equity, social justice)
- (3) transparency and explainability (foster general understanding, make stakeholders aware, enable challenges to AI outcomes)
- (4) robustness, security and safety
- (5) accountability
- Five recommendations for governments: (1) invest in AI R&D
- (2) foster digital ecosystem for AI
- (3) shape enabling policy environment
- (4) build human capacity and prepare for labor-market transformation
- (5) international cooperation
- AI actor categorization: organizations and individuals that play active role in the AI system lifecycle, including those that deploy or operate AI
- AI system definition (updated November 2023): machine-based system that for explicit or implicit objectives infers from input how to generate outputs (predictions, content, recommendations, decisions) that can influence physical or virtual environments
- OECD.AI Policy Observatory: implementation infrastructure with national-policy database, indicators, expert network
- Generative AI considerations (May 2024 update): content provenance and authentication, training-data and intellectual-property, misinformation and societal effects, broader sustainability concerns
Primary use case
Foundational soft-law reference framework for AI policy development globally; intellectual foundation and explicit reference for subsequent AI governance instruments including G20 AI Principles, EU AI Act high-level framing, NIST AI RMF principles, UNESCO Recommendation on the Ethics of AI (2021), AI Bill of Rights; OECD.AI Policy Observatory: comparative AI policy database, live tracker of national AI strategies, expert convening platform; intergovernmental coordination instrument for AI policy harmonization across OECD members and partners; academic and policy reference in AI governance, comparative AI regulation, and international technology policy literature; input to corporate responsible-AI policies citing OECD principles as foundation for internal AI ethics frameworks.
Common criticisms
- The Principles' soft-law status — non-binding recommendations rather than enforceable obligations — has been substantially debated
- civil-society groups including Access Now, Article 19, and AlgorithmWatch have argued the framework relies excessively on AI-actor self-regulation in the absence of binding enforcement, while industry-aligned commentators argue soft-law approaches preserve innovation flexibility
- the high-level abstraction of the principles produces operationalization challenges — what 'human-centered values and fairness' requires in specific deployment contexts is substantially indeterminate, and divergent implementations across signatory countries reflect this
- the framework's consensus-based development required compromise that produced general formulations easier to support than specific commitments, with critics arguing principles like 'transparency and explainability' lack the specificity to drive practice change
- geopolitical critique that OECD-developed principles reflect Western-aligned values and may not adequately represent emerging-economy and Global-South AI governance perspectives, with parallel UNESCO and ITU frameworks offering broader-participation alternatives
- the May 2024 generative-AI update has been argued by some commentators to be reactive rather than proactive, with the original 2019 principles requiring substantial extension for ChatGPT-era technology
- the AI-actor responsibility framework places obligations on developers, providers, and deployers without clear allocation rules for responsibility distribution along the AI value chain
- the OECD.AI Policy Observatory's national-policy tracker documents substantial divergence in implementation approaches across signatories, raising questions about whether shared principles produce shared practice
- interaction with binding regulation (EU AI Act, national AI laws) is conceptual rather than formal, with the Principles serving as policy orientation rather than compliance reference
- cross-border AI governance fragmentation continues despite the Principles' harmonization aspirations.
Lineage
- Siblings
- NIST AI Risk Management Framework, EU AI Act, ISO/IEC 42001, AI Bill of Rights