PolicyLaw, regulation & governance
England's university regulator urges more in-person exams after finding AI can complete unsupervised coursework
England's Office for Students says students can often use AI to complete unsupervised coursework to a high standard without showing the knowledge being tested. It is steering universities toward exams, vivas and presentations, and says weak assessment could breach registration conditions.
From the The Guardian · AI article of November 11, 2026 and cited sources: The OfS and Advance HE study concluded that unsupervised assessments can often be passed with general-purpose AI tools. [3] [1] [6]
SOURCES[3] OfS: Making AI work for students: reflections on… officeforstudents.org.uk[1] AI has made it ‘close to impossible’ to assess university… theguardian.com[6] OfS press release: Universities should help students… officeforstudents.org.uk
AT Pulse research deskPublished · Updated · 5 min read · AI-assisted analysisAI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- The OfS and Advance HE study concluded that unsupervised assessments can often be passed with general-purpose AI tools. [3] [1] [6]
- Original source
- The Guardian · AI article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 11, 2026 article published
- Covered
- Nov 11 edition
- Evidence
- 8 sources across 6 sites
- 1AI has made it ‘close to impossible’ to assess university…theguardian.com
- 2UK universities told to go analogue to stop AI cheatingft.com
- 3OfS: Making AI work for students: reflections on…officeforstudents.org.uk
- 4HEPI: Student Generative AI Survey 2026hepi.ac.uk
- 5OfS: OfS collaborates with Advance HE on AI researchofficeforstudents.org.uk
- 6OfS press release: Universities should help students…officeforstudents.org.uk
Reach: not measured yet
The OfS and Advance HE study concluded that unsupervised assessments can often be passed with general-purpose AI tools. It also found that AI detectors do not reliably identify AI use. The regulator therefore favours controlling where work is produced: supervised exams, vivas, presentations and process-based tasks. [3] [1] [6]
The OfS calls AI a generational risk and will treat unreliable assessment as a compliance issue. Its possible responses include conditions, penalties and suspended funding. The Department for Education asked it to enforce standards quickly, and Universities UK said many institutions have already changed how they assess. [6] [1] [8]
The findings come from roundtables, workshops and interviews with over 300 students and staff at more than 120 institutions. They rest on testimony rather than measured grade effects or detection rates. A separate HEPI survey found 94% of undergraduates use generative AI for assessed work. [3] [7] [4]
students widely use AI for assessed work, and the regulator judges detectors unreliable. Implication: invigilation and staff costs rise.
Read the full assessment
Demand may move from detection tools toward proctoring and oral-assessment tools, and employers may give coursework grades less weight.
Executive brief
England's higher education regulator, the Office for Students (OfS), has told universities that unsupervised coursework can often be completed to a high standard with AI tools, without the student showing the knowledge it is meant to test. The OfS also warned that assessment that fails to measure students reliably could breach registration conditions and lead to sanctions (OfS report; Guardian). One staff member interviewed said AI may make assessment "close to impossible" in some areas. The fix the OfS proposes is not mainly technical: more in-person exams, vivas and presentations, because AI detectors are unreliable.
What changed and event timeline
Student AI use near universal
A HEPI/Kortext survey of 1,054 undergraduates found 94% use generative AI for assessed work, and fewer than half feel supported by teaching staff ().
OfS and Advance HE launch AI study
The plan combined surveys, roundtables and interviews with sector bodies and technology companies, with findings due later in 2026 ().
Fieldwork
More than 300 students and staff from over 120 institutions took part ().
Report and guidance published
The OfS called AI a "generational risk" and said it will require valid, reliable assessment, which in many cases means secure, AI-free formats ().
Government and sector respond
The Department for Education told the OfS to enforce standards "at pace." Universities UK said many institutions have already changed how they assess ().
Capabilities and access
- The report names no specific model or version. It treats general-purpose AI tools as a class.
- The relevant capability is producing work that passes unsupervised assessment.
- HEPI found 95% of undergraduates use AI in some form, so access is effectively universal (HEPI).
Read the full section
- The report names no specific model or version. It treats general-purpose AI tools as a class. The only named product is ChatGPT, mentioned in a student quote (Guardian).
- The relevant capability is producing work that passes unsupervised assessment. Students reported peers finishing assessments "sometimes in just a few minutes" and getting grades similar to those who did not use AI (Guardian).
- HEPI found 95% of undergraduates use AI in some form, so access is effectively universal (HEPI).
Technical analysis for researchers and developers
- Method: The study was qualitative (roundtables, workshops, interviews) and included 91 students or student representatives (Wonkhe).
- Detection: The report concludes that detectors "do not consistently and reliably" identify AI use, and that compliance with AI policy usually cannot be conclusively established (OfS report).
- Design implication: If you can't verify how work was produced, you have to control the setting it is produced in.
Read the full section
- Method: The study was qualitative (roundtables, workshops, interviews) and included 91 students or student representatives (Wonkhe). It did not measure grade effects or detection rates, so its claims about AI and grades rest on testimony.
- Detection: The report concludes that detectors "do not consistently and reliably" identify AI use, and that compliance with AI policy usually cannot be conclusively established (OfS report).
- Design implication: If you can't verify how work was produced, you have to control the setting it is produced in. That points to supervised exams, vivas, process-based assessment (students document how they researched and built the work), and tasks where students critique or edit AI output (Guardian).
- Signals of AI use: Academics reported large numbers of submissions with the same style and the same references.
Claims and evidence
- Unsupervised assessments can be passed with AI. This is the report's conclusion, drawn from participant testimony (OfS report).
- "A significant proportion" of assessments remain unsupervised. This comes from OfS guidance under registration condition B4 (Times Higher Education). No percentage is published.
- Near-universal AI use in assessed work. This is supported independently by a separate HEPI survey, not by the OfS study (HEPI).
Read the full section
- Unsupervised assessments can be passed with AI. This is the report's conclusion, drawn from participant testimony (OfS report).
- "A significant proportion" of assessments remain unsupervised. This comes from OfS guidance under registration condition B4 (Times Higher Education). No percentage is published.
- Near-universal AI use in assessed work. This is supported independently by a separate HEPI survey, not by the OfS study (HEPI).
- "Close to impossible" to assess. This is one staff member's quote about "some areas." It is not a sector-wide finding (Guardian).
- Sample size. Reports differ on the number of institutions: "over 120" (OfS) versus 127 (Guardian).
Context and prior work
- HEPI's 2026 survey found students worried about fairness, loss of independent thinking and being falsely accused of misconduct (HEPI).
- Simon Banks of King's College London told Parliament that some assessment methods are "seriously threatened" (Times Higher Education).
- The Financial Times framed the guidance as telling universities to "go analogue" (FT).
Limitations, safety and contested findings
- Representativeness: The evidence is qualitative and self-selected, so it cannot show how common misconduct is.
- Feasibility: Wonkhe's Debbie McVitty called the guidance both a "baseline expectation and wildly aspirational."
- Positive uses: Some staff see benefits, such as AI-generated feedback freeing up time to spend with students (Times Higher Education).
Read the full section
- Representativeness: The evidence is qualitative and self-selected, so it cannot show how common misconduct is.
- Feasibility: Wonkhe's Debbie McVitty called the guidance both a "baseline expectation and wildly aspirational." She noted institutions face paradigm-level change under financial pressure and without a roadmap (Wonkhe).
- Positive uses: Some staff see benefits, such as AI-generated feedback freeing up time to spend with students (Times Higher Education).
- Unclear rules: Students are often unclear about what AI use their institution permits (OfS report).
Business and practitioner implications
- Enforcement risk: The OfS lists conditions, penalties and suspension of funding as levers it can use (OfS press release).
- Costs: Universities face higher costs for invigilated exams and vivas, which scale with staff time.
- EdTech vendors: The regulator's own evidence weakens the case for AI detection products. Demand is likely to shift toward proctoring, process-capture and oral-assessment tooling.
Read the full section
- Enforcement risk: The OfS lists conditions, penalties and suspension of funding as levers it can use (OfS press release).
- Costs: Universities face higher costs for invigilated exams and vivas, which scale with staff time.
- EdTech vendors: The regulator's own evidence weakens the case for AI detection products. Demand is likely to shift toward proctoring, process-capture and oral-assessment tooling.
- Employers: Coursework-based grades may be read more cautiously. The OfS also expects courses to build AI competencies.
- Sector support: Advance HE plans an AI Change Academy in early 2027 (OfS press release).
The source trail.
Sources (8)
AI has made it ‘close to impossible’ to assess university coursework, regulator says
Article text retrieved; extracted text may omit tables or interactive elements.
theguardian.comUK universities told to go analogue to stop AI cheating
Related coverage; assess separately
ft.com