OECD updates AI Principles for generative AI era

The OECD revised its AI Principles, first adopted in 2019 and among the earliest intergovernmental frameworks for AI policy, to address risks specific to generative systems. The update was adopted by the 38 OECD member countries plus a further group of adherents — 47 in total, including the EU — none of whom are bound by it; the Principles are a recommendation, not a treaty.

The changes reworked the framework’s language rather than its structure. The second principle, “human- centred values and fairness,” became “human rights and democratic values, including fairness and privacy,” and new text addressed AI-amplified misinformation and disinformation, environmental impact, and the need to report safety incidents involving AI systems — none of which had existed as concerns when the original text was drafted.

Because the Principles carry no enforcement mechanism, their significance was as a marker of consensus rather than as regulation: a signal that generative AI’s risks — deepfakes, manipulation, safety incidents — had moved from a lab-safety conversation into the vocabulary of mainstream international policy, ahead of any of the adherent countries passing binding legislation to match.