DfE science advisers set 8 rules for judging AI in schools
The advisers say AI changes faster than full trials can test it. They want evidence gathered all the time and AI suppliers to share usage data with schools.
Scientists who advise the Department for Education published eight principles on 24 September 2026 for deciding how AI should be used in schools.
The report comes from the AI and digital education working group of the department's Science Advisory Council. It is independent advice, not government policy. Its authors include Professor Dame Athene Donald, who chairs the council. The department's chief scientific adviser, Professor Michael Thomas, is another.
The demand for RCT-standard evidence before practitioner guidance is feasible is not a neutral methodological choice.
The group says AI tools change faster than randomised controlled trials can test them. A trial's results may describe a version of a product that no longer exists. It cites a Stanford review of about 800 papers on AI in education, which found only 20 that met its standard for showing cause and effect.
Instead, the group wants evidence built up from many sources and updated as it arrives. This would include reports from teachers. It says caution should come first where there is a credible risk of serious harm to children. For lower-risk uses, such as tools that help teachers with admin, it favours learning as use grows.
Two principles affect schools most directly. One warns against buying decisions that create a dependence on a single supplier before the evidence is in. Another suggests AI suppliers could be required to give schools and independent evaluators usage data as a condition of public sector contracts.
For teachers, the report does not change any rule. It says the principles sit alongside existing department guidance. Inside government, they are meant as a shared reference for decisions about AI in education. For suppliers, they set out the questions about evidence that developers should be ready to answer.
The department has not said whether it will adopt the usage data proposal. Its safety certification plans, published on 10 November, would check AI products for safety but not for whether they help pupils learn.
Sources
- Department for Education (GOV.UK) Evidence-informed policymaking and fast-changing technology
- DfE Science Advisory Council Principles for evidence-informed policymaking in fast-changing technology contexts (PDF)
- Department for Education (GOV.UK) Department for Education establishes Science Advisory Council