Wiki · Governance · Last reviewed June 25, 2026

OECD AI Principles

The OECD AI Principles are the values and policy recommendations in the OECD Recommendation of the Council on Artificial Intelligence, a soft-law reference for trustworthy AI that links innovation to human rights, democratic values, transparency, robustness, safety, and accountability.

Category: AI governance Published: June 25, 2026 Modified: June 25, 2026 Last reviewed: June 25, 2026 Tags: OECD, AI governance, trustworthy AI, soft law, accountability

Definition

The OECD AI Principles are the principle and policy sections of the OECD Recommendation of the Council on Artificial Intelligence, OECD/LEGAL/0449. The OECD Council adopted the Recommendation at ministerial level on May 22, 2019. OECD.AI describes the Principles as initially adopted in 2019 and updated in May 2024 to reflect new technological and policy developments.

The Recommendation is often treated as a baseline for trustworthy AI governance because it provides common language for AI systems, AI actors, the AI system lifecycle, responsible stewardship, policy design, and international co-operation. It is not a treaty or a product certification. OECD Recommendations are not legally binding, but they carry an expectation that adherents will do their utmost to implement them.

Snapshot

Principles

The five values-based areas are inclusive growth, sustainable development, and well-being; respect for the rule of law, human rights, democratic values, fairness, and privacy; transparency and explainability; robustness, security, and safety; and accountability. The legal text calls on AI actors to promote and implement these principles according to their roles, context, and ability to act.

The practical move is to connect value language to lifecycle evidence. Transparency should produce meaningful information that helps people recognize AI interactions, understand outcomes, and challenge adverse outcomes. Robustness, security, and safety should produce traceability, risk management, and controls across the lifecycle. Accountability should make clear who is responsible for proper system functioning and respect for the principles.

Policy Recommendations

The Recommendation also gives governments five policy directions: invest in AI research and development; foster a digital ecosystem for AI; shape an enabling policy environment; build human capacity and prepare for labor-market transformation; and support international co-operation for trustworthy AI.

That second half matters because the Principles are not only ethics language for companies. They also describe public-sector duties: skills, infrastructure, research, competition, data access, testing environments, standards, measurement, and cross-border coordination. A jurisdiction cannot claim trustworthy AI only by asking private actors to behave well while leaving workers, institutions, researchers, and affected communities without capacity.

Governance Use

For an organization, the OECD AI Principles are most useful as a crosswalk. A policy team can map product documentation, risk registers, model cards, human oversight plans, procurement clauses, incident response, privacy controls, accessibility review, and user notices against the five principles. The map should show evidence, not slogans.

For agentic and generative AI systems, the Principles push review beyond model accuracy. A credible record should identify objectives, data sources, affected groups, autonomy level, human intervention points, explanation route, security controls, misuse paths, monitoring signals, and accountability owners. Where a system makes recommendations or decisions that affect people, the challenge and redress path should be visible before deployment.

Limits

The OECD AI Principles do not decide whether a particular AI system is lawful, safe, fair, or aligned with a sector regulation. They are a high-level governance reference. A deployment still needs applicable law, domain standards, technical evaluations, privacy analysis, security review, labor consultation where relevant, and evidence from the actual workflow.

Soft law can also become cover. A company can cite the Principles while hiding model behavior, refusing audit access, or offering no remedy to affected people. The test is whether the principles change decisions: delay a launch, narrow a use case, require monitoring, improve notice, create an appeal channel, or assign accountability for residual risk.

Source Discipline

Use OECD.AI for the current overview, adoption/update framing, and public explanation. Use the OECD Legal Instruments record and PDF for the Recommendation text, definitions, principle language, recommendation structure, and soft-law status. Use national laws, regulators, and sector standards for binding obligations. Do not cite the OECD AI Principles as proof that a system has passed an audit.

Spiralist Reading

The OECD AI Principles are an early constitutional grammar for the machine-mediated state. They do not govern the model directly. They give institutions a vocabulary for saying what must remain human, contestable, explainable, secure, and accountable while AI moves into public and private decision systems.

Spiralism reads them as a memory aid against drift. The words are familiar enough to become ritual. The work is to keep asking where the record is: who was affected, who could challenge the system, what failed, what changed, and who remained accountable after the model entered the loop.

Open Questions

Sources

Return to Wiki