EU AI Act's High-Risk School AI Deadline Moves to 2027, Not Yet Law
The EU's tough rules for AI grading, admissions and exam-monitoring tools were due to bite this week but have been pushed to December 2027 — a delay that isn't yet legally final, leaving schools and vendors in limbo.
Yesterday, 2 August 2026, was supposed to be the day the European Union's toughest AI rules for schools kicked in. They didn't. A last-minute political deal has pushed the deadline back to December 2027 — except that deal isn't law yet, and won't be until the EU Council signs off and it's published in the Official Journal.
For UK schools, MATs and the edtech vendors that supply them, this is not a distant continental squabble. It's a preview of the compliance questions British procurement teams will be asking within the next year, whether or not Westminster ever passes an equivalent law.
What counts as "high-risk" AI in a school
The EU AI Act, in force since August 2024, classifies four categories of educational AI as high-risk under Annex III: systems that determine access or admission to an institution, systems that evaluate learning outcomes (automated grading and marking tools), systems that assess the appropriate level of education a student can access, and systems that monitor students for prohibited behaviour during tests — exam proctoring software, in plain terms.
That list covers a meaningful slice of what schools already buy: AI-assisted marking tools, adaptive placement software, and remote-invigilation platforms used for online assessment. Providers of these systems are supposed to run risk management processes, maintain data governance standards, build in human oversight, log system behaviour, and pass a conformity assessment before the tools go anywhere near a live decision about a student. Non-compliant vendors face fines of up to €15 million or 3 percent of global turnover, whichever is higher.
Why the deadline slipped
According to reporting from Gibson Dunn and other law firms tracking the file, the European Parliament and Council reached a provisional agreement in the so-called Digital Omnibus process to defer the Annex III high-risk obligations from 2 August 2026 to 2 December 2027. The stated reason is that the standards bodies responsible for the technical specifications underpinning conformity assessment have not finished their work, and neither regulators nor vendors were ready to enforce or meet the original deadline.
Crucially, the delay is provisional. Until the Council formally approves the text and it appears in the Official Journal, 2 August 2026 remains the technically binding date. Legal advisers are telling clients to keep building towards compliance rather than assume the reprieve is guaranteed.
The postponement delays enforcement. It does not delay the underlying expectation that AI used to admit, grade or monitor students should be transparent and subject to human oversight.
Why this touches UK schools at all
The UK left the EU, and Westminster has no equivalent statute for classroom AI. But the Act reaches further than EU borders. Any company placing an AI system on the EU market, or whose tool affects people located in the EU, falls within scope — which catches UK-headquartered edtech firms selling into European schools, international schools and sixth-form colleges with EU-resident pupils, exam boards running assessments across borders, and universities recruiting from the continent.
That means the marking assistant, adaptive learning platform or proctoring tool a UK school procures from a vendor with European customers may already be shaped by these rules, whether the school ever hears about it. It also means the compliance conversation vendors are having with EU regulators now is a reasonable proxy for the questions UK governors and IT leads should be asking before they sign the next contract.
What to do
Ask any AI vendor used for grading, admissions screening or exam monitoring whether they classify their tool as high-risk under the EU AI Act and what their compliance timeline looks like, even if your school has no EU exposure — it's a useful proxy for how seriously a vendor takes governance generally.
Build a short internal note on which AI tools in your school touch admissions, grading or exam integrity decisions. These are the categories regulators, in the EU and eventually elsewhere, keep returning to.
Don't treat the delay as a reason to relax. December 2027 is closer than it sounds for procurement cycles that run two or three years, and the requirements — human oversight, transparency, documented risk management — are sound practice regardless of which regulator enforces them.
What to watch
Watch for the Council's formal approval and Official Journal publication, which will make the December 2027 date legally certain rather than provisional. Watch how exam boards and international qualification providers — many of whom operate across UK and EU markets simultaneously — respond, since their compliance posture will likely set a de facto standard well beyond Europe. And watch whether the Department for Education or Ofqual references the EU framework as a model, given the UK currently has no comparable statute governing high-stakes assessment AI.