England threatens universities: bring back in-person exams to stop AI
England's education regulator says AI makes grading coursework impossible. They are forcing universities to bring back in-person exams.
Policy ยท Source: Times Higher Education
What happened
England's higher education regulator is pulling the plug on remote coursework. The Office for Students says AI makes fair grading close to impossible. They are urging universities to return to in-person exams, oral vivas, and practical tests. Institutions that fail to secure their assessments face strict sanctions. These penalties include monetary fines, lower quality awards, and suspended access to student support funding. The message is clear. The era of unsupervised essays is ending.
The crackdown follows a massive joint study with Advance HE. Researchers spoke to over 300 staff and students across 127 institutions. The findings are bleak for traditional homework. Students are using large language models to finish assignments in minutes. They get the same grades as peers who do the work manually. Worse, these students cannot recall basic details of their own submissions. Instructors report receiving huge batches of essays with identical styles and references.
The regulator admits that AI detection tools do not work. Human reviewers cannot reliably spot AI text either. Because detection is a dead end, the only fix is structural. Universities must design assessments that physically prevent AI access. If an assignment can be completed by AI without the student learning the material, it must be scrapped. Simon Banks from King's College London told Parliament that universities must be able to certify students with confidence. That means bringing assessment back into the exam hall.
Key facts
- 127 โ Number of institutions involved in the joint research study
- 300 โ Approximate number of students and staff members interviewed
- 2027 โ Year Advance HE will launch an AI-focused Change Academy
Why it matters
EdTech founders need to pivot fast. Building another AI detection tool is a waste of time. The regulators have officially declared that detection software is unreliable and cannot be used as sole evidence for misconduct. Suspicion without strong evidence will not justify a lower mark. The money is now in secure testing environments. Universities need infrastructure for in-person exams, oral assessments, and controlled digital environments that lock out generative AI completely. The institutions hold the budgets, and they are terrified of losing their funding over compliance failures.
This creates a massive shift in how students interact with AI. Right now, students use AI as a homework machine to bypass productive struggle. Under the new rules, they will have to use AI as a tutor. If the final test is a closed-book exam, offloading the cognitive work to a language model guarantees failure. Tools that help students actually learn and retain information for offline exams will become essential. Students still want AI to explain concepts and generate practice questions. The market for AI tutors is about to explode as the market for AI essay writers collapses.
For builders
Stop building AI text detectors
The Office for Students explicitly stated that detection tools do not consistently identify AI use. Universities cannot use them as hard evidence for academic misconduct. If you sell detection software, your market is evaporating.
Build secure assessment infrastructure
Universities are scrambling to comply with the new mandate for secure testing. They need scalable ways to conduct and grade oral vivas, in-person exams, and practical tasks. Founders who build tools to streamline offline testing logistics will win university contracts.
Shift to AI tutoring platforms
Students can no longer submit AI-generated text for grades. They must perform in the exam hall. There is a massive opportunity for AI tools that test knowledge, generate practice questions, and force active recall.
My take
I have said this from day one. You cannot out-engineer AI with a text detector. Stop fighting the AI and start building systems that actually measure human knowledge in the real world.
Original reporting: Times Higher Education. This is my rewrite and opinion.