England’s university regulator sets a new test for AI-era degrees: Secure, AI-free assessment

England’s university regulator, the Office for Students (OfS), says it will require universities to prove their assessments validly and reliably measure what students know, “often through secure, AI-free assessment”. The commitment comes in a press release published on 11 November 2026 with new research done with Advance HE.

The Guardian was among the first to report the story, saying the regulator considers AI makes marking coursework “close to impossible” and wants a return to traditional exams and in-person presentations.

What the research found about unsupervised work

The OfS and Advance HE describe AI as both an opportunity and a risk. Universities and colleges, they say, have a major role in teaching students AI skills for later careers, but AI also threatens how reliable assessment is and how far qualifications can be trusted.

According to the release, students described classmates who used large language models to finish unsupervised assessments in minutes and received grades similar to those of students who did the work themselves. Staff told the researchers that assessments do not consistently or reliably measure knowledge and skills.

The research concludes that closed-book exams, presentations, practical assessments and performances are likely the most valid measures of attainment, because they are designed to limit AI use. Other findings:

• AI use among teaching and professional services staff varies widely, within and between institutions.
• Some student uses are sophisticated, but both students and staff worry that others skip the effortful thinking deep learning needs.
• Students find institutional AI policies hard to understand, and some feel staff use of AI has lowered the quality of learning materials.
• AI literacy training for staff is limited, generally voluntary and poorly attended.
• Students see AI skills as important for jobs and want more chances to build them.

Fines and funding are on the table

The regulator lists what it will do. It will require institutions to make sure assessments measure knowledge and skills validly and reliably, and to keep courses up to date and skill-building. It says it will hold providers to account with tools including registration conditions, monetary penalties, and suspending student support or OfS grant funding. It will also clarify its requirements, share examples of good practice and carry out more research on the scale and nature of AI’s effects.

Josh Fleming, the OfS director of strategy and delivery, said AI “presents a generational risk to the system as it stands”. Advance HE, the sector body, plans an AI-focused Change Academy for members in early 2027, a convening of sector leaders and academics, and a review of its Knowledge Hub resources on ethical AI use.

A shift from advice to enforcement

The notable change is the enforcement language. Until now, commentary on the regulator’s role has mostly been advocacy. A HEPI piece argued the OfS should make AI assessment policy an auditable condition of registration, and noted that its existing B4 conditions mention essay mills and plagiarism but not generative AI. The release does not say whether a new condition will be created or an existing one reinterpreted.

The scale of the change is large. In August, the think tank Policy Exchange reported that its authors estimated about 78% of the 120 universities they asked relied on online remote exams in 2024, and that 70% planned to keep using unsupervised online exams, according to Times Higher Education. If those figures are close, many providers would have to rebuild how they assess.

England’s qualifications regulator for schools is moving the same way. Ofqual chief Ian Bauckham said written coursework will face “far, far more scrutiny” under reformed qualifications, Results Sense reported.

What the announcement leaves open

The release sets direction, not deadlines. It gives no date for when secure assessment becomes a requirement, no definition of what counts as secure, and no penalty amounts. The finding that exams and presentations are the most valid measures is the researchers’ conclusion from student and staff accounts, not a measured comparison of grades. The Guardian’s headline phrase about marking being near impossible is the newspaper’s characterization, and it does not appear in the parts of the release read here.

For students, the likely effect is more timed, supervised and spoken assessment. The OfS also says institutions should keep teaching AI skills, so the message is not a ban on AI. The full report is titled Making AI work for students: reflections on institutional practice, and the OfS is holding a webinar on next steps for the sector.