England’s higher-education regulator has urged universities to use assessments that can more securely demonstrate students’ learning, citing concerns that students’ access to artificial intelligence tools could undermine confidence in some coursework. The guidance does not require institutions to reinstate traditional exams across the board.
The Office for Students (OfS) said assessments that limit access to AI will often be appropriate, but it also allows institutions to permit AI when they clearly explain the rules. It listed in-person exams, presentations, oral examinations, performances and practical tasks as possible ways to assess learning.
A call for secure, valid assessment
The guidance follows research involving the OfS and Advance HE, according to a report by Times Higher Education. Over four months, the research drew on conversations with about 300 students and staff from more than 120 institutions. The report described participants’ accounts of peers using AI in assessments, sometimes quickly and while receiving similar grades.
Those accounts indicate concerns among people involved in higher education, but they are not a measure of how widespread undisclosed AI use is across all universities. The same report said a significant proportion of assessments remain unsupervised, leaving students able to access AI tools.
The concern is about whether an assessment can reliably show what a student knows and can do. The reporting supports a risk to assessment validity, but it does not establish that coursework as a whole has become impossible to assess. The phrase “close to impossible” should therefore be treated as headline wording, not as a confirmed formal finding or exact statement by the OfS.
The regulator’s examples point to a range of possible responses rather than a single prescribed format. Institutions can choose assessments suited to their courses and learning aims, while setting out when students may use AI and when they may not. The guidance does not prohibit coursework or the use of AI in every assessment.
Regulatory expectations, not an automatic penalty
The OfS framed its guidance in relation to existing expectations for assessment quality, including condition B4. Providers that fall short may face scrutiny under the regulator’s quality framework. The reporting described possible lower outcomes or further investigation, not an automatic sanction for offering coursework or allowing AI use.
That distinction matters for institutions deciding how to respond. The regulator’s message, as reported, is to ensure that assessment remains a credible measure of learning. It is not an order for every university to abandon take-home work or return to a single model of supervised examinations.
The guidance also leaves room for assessments in which AI use is part of the task, provided the rules are made clear. That approach reflects the difference between undisclosed assistance that obscures a student’s own work and permitted use that an institution has explicitly built into an assessment.
Parliamentary evidence points to a broader debate
The OfS guidance came amid wider discussion about how universities should assess learning as generative AI becomes more relevant to study and work. In parliamentary evidence on 10 November 2026, Simon Banks, a professor at King’s College London, said generative AI threatened some traditional assessment practices. He argued for a mixed portfolio that could include exams, oral assessments and practical work, designed inclusively.
Other witnesses in the hearing said assessment should reflect what students are meant to learn. They also raised the possibility that students’ ability to use AI could itself be assessed. That evidence supports the broader case for adapting assessment, but it does not establish the precise wording or requirements of the OfS guidance.
Witnesses also raised concerns about fairness in more controlled assessment. A student representative said allegations of AI misconduct might disproportionately affect international students, including those using translation or writing-support tools. Another witness warned that remote-proctoring systems could falsely flag disabled students. These were concerns raised in testimony, not findings independently verified by the hearing.
Such concerns are relevant when universities consider replacing take-home work with supervised or monitored formats. The evidence does not show that controlled assessment is inherently unfair, but it does point to the need to consider how assessment rules and monitoring systems affect different students.
Different regulators, related questions
Ofqual, which regulates qualifications rather than university degrees, has separately described supervised exams as relatively protected from student access to AI, while coursework and other non-exam assessments face greater pressure. Its approach emphasizes validity, fairness, security and innovation rather than treating every use of AI as misconduct.
That provides context for the assessment debate, but Ofqual’s remit is distinct from the OfS’s role in higher education in England. Advance HE has confirmed its partnership with the OfS on research into AI’s opportunities and risks in higher education, including academic integrity and access and participation. That announcement confirms the research partnership, rather than independently verifying the November findings.
