Your university almost certainly has a position on you using AI to write an essay. The question is whether you can find it, whether it means what it says, and whether it was written to help you or to catch you. A new Higher Education Policy Institute report just read every one it could find, and the answer is less reassuring than the friendly intro paragraphs suggest.
The report — "What UK university AI policies actually do: A study of 96 institutions" — was written by Sam Illingworth, professor of creative pedagogies at Edinburgh Napier University, and published by HEPI on 21 May 2026. He didn't survey universities and ask them to describe their own policy. He scraped the public web, on a single day, for whatever a student, parent or regulator could actually find without logging in — which turns out to matter as much as what the policies say.
The auditTwo in five couldn't even be found
Illingworth checked all 163 UK higher education institutions. Just 96 had an AI policy that was publicly accessible — no student portal login required, no "ask your department" runaround.
Of the 96 he could read, the computational pass classified the large majority — 83 of them — as written in the language of education: critical literacy, skills, learning outcomes. That sounds like good news. It isn't, quite.
The catchEducational language, compliance machinery
Illingworth's core finding is a gap between how these policies talk and what they actually do. His description, as reported by Times Higher Education: many policies "use the language of education while operating as compliance instruments." The University of Southampton's policy is his example — it promises to produce "critically digitally literate" graduates, but reads in practice as a binary list of what's allowed and what isn't, opened with a threat:
"We will take disciplinary action that may result in penalties on your marks."
His read on why this keeps happening isn't that universities are acting in bad faith. It's structural: AI guidance got written into misconduct and governance frameworks that already existed, and those frameworks were already built around detection and punishment, not teaching. As he puts it, "the response to AI has been absorbed into existing governance structures" — so a document that opens with "here's how to use this well" closes with the same penalty clause as a plagiarism policy, because it was drafted inside one.
The other structural problem: there's no sector body coordinating any of this. Every institution built its policy alone, which is why the 96 range from genuine critical-literacy resources all the way down to a single paragraph bolted onto an existing misconduct procedure.
Search "[your university] generative AI policy" and read whatever comes up — properly, not skimmed. Note three things: does it require you to disclose AI use, does it name specific banned tools or tasks, and does it mention detection software.
If you can't find anything official, that's an answer too. Email a tutor and ask directly what's allowed for your next assignment, in writing, before you submit — not after you're asked to explain yourself.
Worth copyingThe four that got it right
Illingworth names four policies that stand up to scrutiny: Durham University, the University of Stirling, Canterbury Christ Church University, and Arts University Plymouth. What sets them apart isn't that they're more permissive — it's that they're built around disclosure and judgement rather than a banned-tools list. Stirling, for instance, uses a graded assessment scale rather than a flat yes/no, so the policy can flex per task instead of per institution-wide decree.
- Read your policy before you need it. The worst time to learn what your university actually requires is mid-appeal.
- Treat "no public policy" as a gap to close in writing. A tutor's emailed reply is the policy you'll actually be able to point to later.
- Notice which kind of document you've got. A one-paragraph addition to misconduct rules reads very differently from a scale or a framework — and tells you how much benefit of the doubt you're likely to get.
None of this means every university is quietly building a surveillance system. It means the paperwork most students never read was mostly written for the moment something goes wrong, not for the moment you're deciding, honestly, how much help is too much. Read it on a quiet afternoon instead of a stressful one.
Get the next one in your inbox
One AI tool, one prompt, three quick wins, every week. Free, and under 3 minutes to read.
Subscribe Free →