AI-free assessment is now England's test of what a degree is worth

England's higher education regulator is pushing universities toward far greater use of AI-free assessment, because its quality rules require every degree to rest on valid, reliable assessment. On 11 November 2026 the Office for Students (OfS) published its guidance on artificial intelligence in higher education, calling for more in-person exams, oral exams, presentations and practical tasks wherever AI use has to be prevented. For universities in England, AI and university assessment now sit inside the regulator's quality rules, and the same question is on its way to every campus that awards a degree.

Watch any undergraduate at 1 a.m. with an essay due at 9. The tool that could write the whole thing is one tab away, and most students now use it. The Higher Education Policy Institute's Student Generative AI Survey 2025, based on 1,041 undergraduates surveyed in December 2024, found that the share using generative AI for assessments jumped from 53% to 88% in a single year. Those students are the reason this guidance exists. They are also the people it protects, because a degree that cannot vouch for its holder loses value on the day they graduate.

What the Office for Students asked universities to do

The guidance draws on research the OfS carried out with Advance HE, which reached more than 300 students and staff from 127 institutions, according to The Telegraph via Yahoo. Unite.AI noted that participants chose to take part, so the findings read best as a picture of sector opinion, with the limits any self-selected group carries. Josh Fleming, the OfS director of strategy and delivery, described AI as a generational risk to the higher education system as it stands and also as a large opportunity, and he called for far greater use of AI-free assessment.

Where AI use must be prevented, universities should use more secure formats, including in-person and closed-book exams, oral exams and vivas, presentations, and practical tasks. Any assessment that AI can complete without the student having the intended knowledge or skills should be redesigned, and courses should set clear rules on permitted AI use. AI detectors get a modest role: their output counts as one piece of evidence at most, and suspicion alone is insufficient grounds to lower a mark. The OfS also left space for AI inside assessment. Where a course permits it, the work still has to show the student's own knowledge and skills.

The regulator is working through rules it already has

The OfS is applying its existing quality conditions, which require credible awards resting on valid, reliable assessment, and it has created no separate AI condition to do it. That is sound regulatory design, and it deserves credit. It ties AI to the question the regulator already asks of every course, so institutions can answer it with the quality processes they have spent years building. The reports note that the sanctions available under the OfS regime include monetary penalties and suspension of access to OfS grant funding, and the Department for Education said it expects universities to take the issue seriously and the OfS to enforce existing standards.

Assessment is the term the four-year bet rests on

Every degree is a four-year bet, a wager placed in the institution's name on a future that keeps moving while the student is enrolled. The student puts in tuition and 4 years of their life, and the institution puts in its reputation. Assessment is the term the whole wager rests on. A transcript is a promise to a stranger that this graduate can do what the course claims to teach. When a tool can complete the assessment without the learning it was built to certify, that promise has nothing behind it.

A transcript is a promise to a stranger that this graduate can do what the course claims to teach.

Curriculum half-life raises the stakes. In fields where tools and methods turn over every few years, some of what a first-year student learns will be dated by graduation, a pattern that also shows in the rise of AI majors in US computer science departments. Assessment has to certify the part that lasts, which is the reasoning behind an answer and the ability to defend it in front of someone who knows the field. An oral exam tests that. So does a practical task done under observation.

Plato asked the same question about writing

In Plato's Phaedrus, written in the 4th century BC, Socrates tells the story of Thamus, an Egyptian king who warns the inventor of writing that the new art will give learners the appearance of wisdom without the substance. The worry was about evidence: how a teacher could tell understanding apart from words copied off a page. Writing stayed, and that question became a permanent part of teaching. The OfS guidance belongs in the same line. It accepts the tool and asks for proof of the person.

The cost side deserves a fair hearing

Participants in the OfS research raised real costs. Exams and oral tests demand staff time and money, and some participants doubted that a closed-book exam reflects professional practice, where references and tools sit on every desk. Others warned that a shift toward exams and vivas could disadvantage disabled students, a serious concern for any institution with access commitments.

These concerns belong in the design conversation from the first draft. A viva for 400 first-year students is a staffing problem, while a 10-minute oral defense attached to a capstone project is a manageable one, and accommodations planned at the start cost less than appeals handled at the end. Academics outside England are working through the same academic-integrity questions, as a recent survey of academics in Ireland showed.

What this means for American provosts and trustees

This guidance applies to England, and US institutions answer to their own accreditors and state rules. The question underneath it travels well. Assessment is the term a degree rests on, and a regulator has now said so plainly in the context of AI. Provosts, deans, CIOs and trustees in the US will meet the same question from accreditors and from their own faculty, and the institutions that have already mapped which of their assessments AI can complete will have the easier conversation. Campuses and accreditors can work through this together, and the OfS has given both a clear, public text to start from.

Here is the question I would take into your next assessment-policy meeting. Pick the single assessment your flagship program leans on most, and ask what its grade would tell an employer if a student handed the task to an AI tool tonight. If nobody at the table can answer with confidence, that assessment is where your four-year bet is exposed. The four-year bet, and how fast course content expires inside it, are questions I bring to my work as Innovator-in-Residence at Tulane University. I'm easy to find.

Here is what makes Alex a credible voice on this topic: As Innovator-in-Residence at Tulane University, Alex works inside the kind of institution this guidance addresses, where the value of a degree and the way it gets assessed are live questions.

Talk with Alex about assessment and the four-year bet →