Skip to content
The Journal

AI at UK universities in 2026/27: what you can and cannot do

UK universities set their own AI rules, the regulator wants more in-person exams and detectors are unreliable. How to stay compliant and still use AI to learn.

The ProoV Team··5 min read

There is no single UK rule on using AI in coursework. Each university sets its own policy, and many modules set their own rules inside that. The Russell Group, the group of leading research universities, published five shared principles in July 2023. They say students should learn to use AI well, that staff should help them, and that teaching and assessment should adapt. They do not say what you may do in a given essay.

The 2026 numbers show why the rules matter. The Higher Education Policy Institute, called HEPI, runs a yearly survey of UK undergraduates. Times Higher Education reported that the 2026 edition found 95% use AI in some way and 12% put AI-generated text directly into assessed work, up from 3% in 2024.

What the regulator said on 11 November 2026

The Office for Students, the regulator for universities in England, and Advance HE published a report called Making AI work for students on 11 November 2026. You can read it on the OfS site. Three points matter to you.

First, work done without supervision can often be done to a high standard with AI. Second, AI detectors do not consistently and reliably identify AI use, and students find their university's policies unclear. Third, the OfS wants far more assessment that does not allow AI: closed-book exams, vivas, presentations and practical tests. A viva is a spoken exam where you answer questions about your own work.

The regulator also said it can act on this through its conditions of registration, fines and the suspension of funding. Expect universities to move toward in-person checks of what you know.

What university policies look like

A HEPI study cited by Times Higher Education found 96 of 163 universities, 59%, had a public AI policy. The most common design is a traffic light: green for assessments where AI is allowed, amber for limited use, red for none. Bath moves to a two-lane model from 2026/27. In its open lane, using generative AI is not misconduct in itself, but plagiarism and invented sources still are. Cardiff, Durham and Edinburgh Napier require students to declare AI use.

Online exams are also still common. A freedom of information study found 78% of UK universities still use some. That means the trend is toward in-person checks, but it has not arrived everywhere.

How common are misconduct cases

A Guardian freedom of information request, reported in June 2025, found about 7,000 proven cases of AI misconduct in 2023-24. That is 5.1 per 1,000 students, up from 1.6 the year before. More than a quarter of universities did not record AI cases separately, so the true number is uncertain. A separate Times request covered Russell Group universities in 2024/25 and counted 2,053 undergraduates punished. The two use different universities and different definitions, so do not read them as one trend.

Employers watch this as well. ISE's October 2026 survey found 92% of employers suspect or have confirmed misconduct in their selection process, and 82% now give candidates guidance on AI.

A checklist for staying compliant

  1. Read the brief of each assessment for the AI rule. Do not assume last term's rule still applies.
  2. Declare your AI use whenever the module asks you to, and say exactly what you used it for.
  3. Keep your notes, drafts and prompts. If a tutor asks how you wrote something, you can show it.
  4. Assume a viva or an in-class check may follow, and make sure you can explain every argument in your own words.
  5. Do not trust a detector score in either direction. A low score does not clear you and a high score does not convict you.
Quick check

A module brief is silent on AI. What is the safest step before you use it for a graded essay?

Use AI to learn without outsourcing the thinking

The students who gain most use it to test their understanding: asking for counter-arguments, explaining a concept three ways, or checking their own reasoning. The safe standard is the one a viva will test. If you could not explain the work aloud, you did not do enough of it yourself.

The same habit is what employers want. The ISE data on graduate jobs is in UK graduate jobs in 2027. A finished project you can talk through is the strongest answer, and proof of skills in hiring in 2026 explains why.

From ProoV

Challenges to prove it

Stop reading, start building. Every challenge runs on demo data and ends in a verifiable certificate.

See all projects