Governing AI in Education: Uneven Rules, Thin Evidence and Procurement as Policy, 2026

During 2026, the governance of AI in education changed faster than in any previous year. Intergovernmental bodies issued their strongest state‐ ments yet: the Council of the European Union adopted its first conclu‐ sions on teachers and AI, UNESCO, UNICEF and ITU launched a Charter for Public Digital Learning Platforms, and education ministers meeting at UNESCO called for education to remain a common good in the age of AI. Binding rules, however, moved in opposite directions. England placed AI inside its statutory safeguarding guidance for schools, while the European Union postponed the date from which its AI Act obligations for high-risk educational systems apply, from August 2026 to December 2027. Mean‐ while, the strongest review of the evidence found only about 20 high- quality causal studies among more than 800, and schools continued to buy AI tools largely on promise. This article analyses these developments, drawing on a practitioner newsletter series written for school leaders over the same period. It argues that the most consequential governance of AI in education is currently happening by default, through institutional procurement, and that this shifts decisions about data, pedagogy, cost and environmental impact to contracts that few people read. It proposes a five-layer governance stack, identifies the gaps between layers, and sets out a procurement-as-governance protocol that schools can apply without waiting for regulation.