The OECD revised its AI principles five years after adopting them

The design

The Organisation for Economic Co-operation and Development first adopted its AI Principles on 22 May 2019, then revised them on 3 May 2024. The instrument's own record on the OECD Legal Instruments database lists it under its formal name, the Recommendation of the Council on Artificial Intelligence, OECD/LEGAL/0449, and records both the 2019 adoption date and the 2024 amendment date. The OECD's own AI Principles overview describes the update as addressing technological and policy developments since 2019 while keeping the same structure: five values-based principles for trustworthy AI, covering human rights and fairness through accountability, plus five recommendations to policymakers on research, ecosystems, and international cooperation.

What the evidence says

Both documents agree on a fact with real weight for how the OECD Principles function: the instrument's own database classifies it as a Recommendation of the Council, the OECD's non-binding legal instrument type, not a Decision, which under OECD procedure would bind adhering governments. The overview page states plainly that the Principles operate as guidance rather than enforceable law, while noting that 47 countries plus the European Union adhere to them and that the EU, the Council of Europe, and other bodies have since borrowed the OECD's definition of an "AI system" and its lifecycle framing for their own, separately binding instruments.

What it asks of people

Because the Principles are addressed to governments and to AI developers in general terms, they ask nothing directly of a person using a companion app. What they ask of an adhering government is to promote, in national policy, values such as transparency, explainability, and human oversight of AI systems; what they ask of companies who reference them voluntarily is to treat those same values as design goals, not compliance checkpoints, since there is no OECD enforcement mechanism behind them. A company's claim to follow the OECD Principles is a statement about its own aspirations, not evidence of an external audit.

Privacy and safeguards

The Principles name privacy as one of the values a trustworthy AI system should respect, alongside fairness and human rights more broadly, but the instrument does not itself specify retention limits, consent mechanisms, or age assurance for any category of product, including conversational companions. Its accountability principle asks that organizations be answerable for the proper functioning of AI systems, again without a defined enforcement path. Where a national law separately incorporates OECD-style language, as several AI statutes now do, the binding force comes from that law, not from the Recommendation itself.

• Does a company's reference to the "OECD AI Principles" point to a specific commitment, or only to the Principles' existence?
• Has the reader's own country adhered to the 2024 update, and if so, has it translated any principle into enforceable law?
• Where a law borrows the OECD's definition of an AI system, does it also borrow the Principles' non-binding character, or convert the language into a duty?

The 2024 update matters chiefly as a sign that a widely adhered-to reference document is still being revised five years on, and that its language keeps migrating into other bodies' binding texts even though the Recommendation itself remains voluntary.