OECD AI Due Diligence
OECD Due Diligence Guidance for Responsible AI
Non-binding OECD guidance applying the OECD's six-step responsible business conduct (RBC) due-diligence process to enterprises across the AI value chain, providing practical recommendations for identifying, preventing, mitigating, and accounting for adverse human-rights and societal impacts of AI systems.
Jurisdiction
International
Enacted
Feb 19, 2026
Effective
Feb 19, 2026
Enforcement
Voluntary; aligned with OECD National Contact Points for Responsible Business Conduct that handle complaints under the OECD Guidelines for Multinational Enterprises.
Published by the OECD on February 19, 2026 as a non-binding due-diligence framework operationalising the OECD AI Principles and the OECD Guidelines for Multinational Enterprises in the AI value chain.
OECD - Due Diligence Guidance for Responsible AI (Feb 19, 2026)Why It Matters
Provides the most detailed multilateral guidance to date on translating high-level AI principles into operational due-diligence steps for enterprises, with influence on how National Contact Points evaluate complaints against multinational AI developers and deployers and how downstream procurement decisions in OECD member states reference responsible AI obligations.
Recent Developments
Published February 19, 2026, operationalising the 2024 update to the OECD AI Principles and providing the first OECD-specific application of its responsible business conduct due-diligence methodology to AI systems. Treated by OECD National Contact Points as a reference for handling specific instances under the OECD Guidelines for Multinational Enterprises.
At a Glance
Harms addressed
Who Must Comply
- Enterprises developing AI systems
- Enterprises deploying AI systems
- Enterprises participating in AI supply chains and value chains
- Multinational enterprises subject to the OECD Guidelines for Multinational Enterprises
Safety Provisions
- Six-step RBC due-diligence process: embed responsible conduct in policy and management systems; identify and assess actual and potential adverse impacts; cease, prevent, or mitigate adverse impacts; track implementation and results; communicate how impacts are addressed; provide for or cooperate in remediation
- Specific guidance on impacts associated with general-purpose and generative AI
- Information integrity and privacy guidance reflecting the 2024 OECD AI Principles update
- Cross-sector applicability for developers, deployers, and downstream users of AI systems
- Recommendations on supply-chain due diligence and AI value-chain transparency
View on map
International
Focus Areas
Cite This
APA
International. (2026). OECD Due Diligence Guidance for Responsible AI.
Related Regulations
CoE AI Convention
First legally binding international AI treaty. Signed by EU, UK, US (Sep 2024), Canada, Japan (Feb 2025) and others. Requires risk/impact assessments, transparency, accountability. National security exemptions apply.
UN/ITU AI & Child Rights Statement
Non-binding multilateral statement signed by thirteen UN and international organisations setting out principles for protecting children's rights in the design, deployment, and governance of AI systems, including provisions on harmful content, age assurance, transparency, and child-rights impact assessments.
UN AI Resolution
First-ever UN General Assembly resolution on AI. Adopted unanimously with 125 co-sponsors (US-led). Establishes human rights as applicable to AI lifecycle, encourages regulatory frameworks, and calls for bridging AI divides between countries. Non-binding but sets global normative expectations.
Italy D.Lgs. 179/2026
Second of the two legislative decrees implementing Italy's AI law (Law 132/2025). It designates the national authorities under the EU AI Act, sets national administrative fines for AI Act breaches, creates the Italian AI regulatory sandbox, and adds rules on AI in schools, work, the professions, health and public administration, including a bar on employment decisions taken solely by automated processing.
CA AB 2246
Repeals and replaces the California Age-Appropriate Design Code Act. Requires a business providing an online service, product or feature likely to be accessed by children to estimate the age of child users with a reasonable level of certainty or extend child protections to all users, to configure high-privacy defaults, and to take reasonable steps to prevent specified harms including severe psychological harm. Removes the exceptions to the data-collection prohibitions, lets children void contracts entered through manipulative design features, and states that it creates no private right of action.
CA SB 503
Requires developers and deployers of AI clinical decision support systems used by health facilities, clinics and physician practices to make reasonable efforts to identify systems with a known or reasonably foreseeable risk of biased impacts on people with protected characteristics. Developers must mitigate those risks and give deployers an intended-use and risk statement plus documentation on training data, evaluation and data governance; deployers must regularly monitor the systems and mitigate.
Last updated May 10, 2026. Verify against primary sources before relying on this information.