England’s university regulator pushes AI-free tests and AI skills

These qualitative accounts identify weaknesses in assessment design. They do not establish what proportion of submissions involve misconduct or measure the effectiveness of reforms across the sector.

The report describes students using AI to explain difficult concepts, generate practice questions and support revision. It identifies potential benefits for disabled and neurodivergent students and those whose first language is not English. But it warns that delegating reading, research and problem-solving can remove the intellectual effort through which students develop knowledge.

An assessment completed within an institution’s AI policy must still establish that its learning objectives have been met. The report says detection tools do not consistently and reliably identify AI use, while students often struggle to understand what their institutions permit.

A separate March 2026 survey from the Higher Education Policy Institute, an independent UK higher education policy organisation, helps distinguish different uses. Conducted by Savanta in December 2025 among 1,054 full-time UK undergraduates, it found that 94% reported using generative AI to help with assessed work. A much smaller 12% reported including AI-generated text directly in assessed work.

Neither figure is a cheating rate: the survey publication notes that some students are encouraged or required to use AI in assessments. Nor do these UK-wide, self-reported responses measure the prevalence of misconduct in England. The survey was sponsored by Kortext, which supplies AI-powered study and content discovery products to higher education institutions.

There is also evidence that assessment reform is already under way. In the same survey, 65% said assessment had changed significantly in response to AI. Universities UK, which represents 147 higher education institutions, said in its response to the findings that many had made substantial changes, including more in-person exams, presentations, oral examinations and controlled use of AI. The OfS also described alternatives such as students critiquing AI output or explaining how they researched and produced their work.

The government’s Department for Education said it expected universities to take the issue seriously and the OfS to enforce existing standards while making its regulatory framework explicit and durable on AI. The immediate question is how to judge whether institutions’ changes work.

Existing rules allow more than one assessment model

The OfS’s new guidance explains how existing registration conditions apply. Condition B1 covers up-to-date courses and relevant skills; B2 covers resources and support; B4 requires effective assessment and credible awards. Research participants generally warned that new AI-specific registration conditions could constrain innovation and instead sought clarity about existing requirements.

Under B4, the regulator identifies secure assessments as an approach likely to comply where they are necessary to measure intended learning outcomes. It defines these as assessments designed to prevent AI use. Examples include in-person examinations, presentations, vivas—oral examinations—performances and practical tasks.

This stops short of a universal AI-free assessment requirement. Where AI is allowed, students must still demonstrate the intended knowledge and skills, which can include effective AI use. Institutions retain a choice of methods; the test is whether those methods reliably establish attainment.

The guidance also says universities and colleges have a legal duty not to discriminate and may need reasonable adjustments for disabled students, including changes to assessment methods. Selectively provided AI or digital tools must preserve assessment validity without giving an unfair advantage. Accessibility therefore belongs in the design of controlled assessment.

Misconduct proceedings should draw fairly on a range of evidence rather than give disproportionate weight to detection software. Clear, accessible explanations of permitted AI use are also part of the regulator’s examples of compliant practice.

HEPI and Savanta’s survey records students’ views on AI skills and institutional support. These are separate self-reported responses, rather than an audit of university provision. The survey was sponsored by Kortext. Source: Higher Education Policy Institute.

Teaching AI requires resources and oversight

The obligations extend to what universities teach and how they support it. Institutions should consider AI developments in students’ disciplines, skills needed in relevant industries and the ability to evaluate AI outputs critically. They should mitigate disparities in access to tools, including paid versions, and training.

The HEPI survey illustrates the support gap: 68% of respondents considered AI skills essential to thrive in today’s world, while 48% felt teaching staff were helping them develop those skills for future careers. Only 38% said their institution provided AI tools. These are students’ reported experiences, rather than an audit of institutions’ provision.

Universities must also oversee their own AI use. The joint research found that some students felt staff use of AI had reduced the quality of learning materials. In accounts from the research, students were often critical of AI marking and feedback, while staff described AI-generated feedback as a useful starting point that could free time for direct interaction. The guidance calls for oversight to ensure feedback is accurate, specific to the student’s work and aligned with assessment criteria.

The research also cautions against assuming rapid efficiency gains. Most participating staff felt AI had not yet delivered significant gains. They cited extensive oversight of outputs, limited staff capability, immature data infrastructure and slow approval processes as barriers to systematic adoption.

Staff training was frequently optional and poorly attended, while many students reported limited training or none. Relatively few participating institutions had embedded a strategic approach across their operations. The report calls for leadership that coordinates institutional adoption while allowing staff to keep experimenting.

A sector analysis argues that education leaders are trying to secure academic quality while allowing for technological change, often without a roadmap or much money to invest. That practical constraint makes guidance on implementation particularly consequential; setting expectations alone does not demonstrate how institutions can meet them.

What universities must demonstrate next

Under the proposed future quality system described in the guidance, the OfS would expect institutions to demonstrate robust assessment arrangements in provider submissions. They could show that assessments prevent AI use, or that assessments allowing it cannot be successfully completed without demonstrating the knowledge and skills being tested. Concerns could contribute to a lower award in that quality system or further investigation. This remains a proposed process, distinct from the regulator’s existing powers.

The guidance also says the OfS will consider specific action when intelligence raises concerns outside a quality assessment cycle. Institutions therefore need to address assessment validity under current requirements, rather than wait for the proposed system.

Advance HE plans an AI-focused Change Academy for member institutions early in 2027, using peer learning to develop institutional strategies. The OfS promises further structured research into the scale and nature of change.

The evidence still needed is whether redesigned assessments demonstrate students’ attainment while preserving accessibility and developing useful AI skills. More supervised exams are one possible response. Their value, like that of assessments incorporating AI, depends on what they establish about learning.

Finland clears Tesla Assisted Driving, keeps EU vote open

Finland has recognised Tesla’s provisional Dutch approval after changes to branding and speed controls. Drivers remain responsible, and the national decision does not commit Finland to backing EU-wide authorisation.