England’s Regulator Wants AI-Free Exams as AI Use Hits 94%

The Office for Students, which regulates higher education in England, said on 11 November that it will “Require institutions to ensure assessments are valid and reliable measures of students’ knowledge and skills”, adding that “In many cases, this will require the use of secure (AI-free) assessments”, according to its release on a joint research report with Advance HE.

The report says the share of students who report using generative AI tools to help with assessments rose from 53 per cent in late 2023 to 94 per cent in late 2025, citing surveys by the Higher Education Policy Institute. That is a rise of 41 percentage points in two years, by our arithmetic.

Business Pill · Verification Cost

An evergreen Business Engineer explainer on verification cost: cheap to make is not cheap to check. It teaches the general idea only and does not discuss the OfS report.

What the Regulator Said

The release says many institutions “continue to use unsupervised assessments that can be completed to a high standard using AI tools without substantive engagement from students”. Students in the research described peers who completed assessments with large language models, sometimes in a few minutes, and received a similar grade to those who did the work themselves, the release says.

Josh Fleming, Director of Strategy and Delivery at the OfS, said AI “presents a generational risk to the system as it stands, threatening to undermine students’ learning and institutions’ ability to assess them”. He added: “We need to see far greater efforts to tackle these risks, including far greater use of AI-free assessment”.

The report lists what it counts as secure assessments: in-person written exams, presentations or vivas, short discussions to test students’ understanding of their submitted work, and practical or performance-based assessments. It says: “If it becomes possible to complete an assessment using AI without the knowledge and skills being tested, the assessment should be replaced”.

The Enforcement Levers

The release says that where evidence suggests institutions have not met its requirements, the OfS deploys levers that include imposing specific conditions of registration, monetary penalties, or suspending access to student support funding and/or OfS grant funding. Alongside the report, the OfS published how institutions can comply with its regulation in the context of AI.

OfS regulation applies only to English higher education, the report notes. Advance HE said it will launch an AI-focused Change Academy programme for member institutions early in 2027.

The Cost and Detection Angle

The report says participants cited resourcing as a barrier, with in-person exams and vivas described as particularly intensive in resources. Participants also said that on many courses assessments have not been redesigned and remain in forms open to AI misuse, such as open-book essays or problem sets.

On detection, the report says many staff said their institutions use AI detection tools despite research suggesting reliability issues. It says a Turnitin spokesperson told the researchers “that its tool’s output should be used as only one data point when working to understand how work was authored”. The report concludes that “The output of detection tools should be, at most, one piece of evidence” in a misconduct case.

The report also says “many students now see AI as the most effective route to a high grade”. For how AI vendors approach schools, see our piece on Google’s Gemini opt-in for K-12 students.

Our Numbers

The views of 244 higher education staff and 91 students and student representatives from at least 127 institutions informed the work, the report says. That is 335 people, about 73% of them staff, by our arithmetic. The report notes participants were self-selecting and may not be fully representative.

The report separately says surveys suggest between 74 and 94 per cent of students now use AI to help them complete academic work. The 94 per cent figure for assessments is about 1.8 times the 53 per cent of late 2023, by our arithmetic.

The Structural Read

As we read the release, the business lever is cost: the report says participants cited resourcing as a barrier, with in-person exams and vivas described as particularly intensive in resources, so a shift to secure assessment points spending toward exactly those formats.

As we read the report, it narrows the role of detection software: it says detector output should be at most one piece of evidence, and Turnitin told the researchers its output should be only one data point.

Three Implications

FOR UNIVERSITIES The release says the OfS levers include specific conditions of registration, monetary penalties, or suspending access to student support funding and/or OfS grant funding.

FOR DETECTION VENDORS The report says some institutions had stopped using AI detection tools because of concerns about their efficacy.

FOR STUDENTS Many students find their institution’s or department’s policies on AI difficult to understand, the OfS release says.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Agentic Harness War.

The framework argues: “Judgment becomes the highest-value layer of work.”

As we read it, the OfS report applies the same logic to degrees: when an essay can be produced by a model, the regulator wants assessment that shows the student’s own knowledge and skills. The framework predates the report and does not discuss it.

What Is Not Established

The release and report we read set no deadline and no fixed share of assessments that must be secure; the release says the requirement is valid and reliable assessment, which in many cases means secure assessments, it says. Neither names paper exams as a mandate. The report gives no count of AI-related misconduct cases; it says research suggests a significant increase up to the 2024-25 academic year.

The report’s own findings come from roundtables, interviews and online prompts, not a representative survey; the 53 and 94 per cent figures are from HEPI surveys it cites. For an earlier education-sector AI deal, see our piece on Anthropic’s teachers union pact.

The Bottom Line

England’s Office for Students says it will require valid and reliable assessment, in many cases secure (AI-free) assessment such as in-person exams, vivas and presentations. Its levers run to funding suspension. Its report with Advance HE cites student AI use for assessments of 94 per cent in late 2025, up from 53 per cent in late 2023.