In a lecture hall at Queen’s University Belfast, students file out after an oral defence of their latest assignment. No software has scanned their words for hidden machine assistance. Judy Williams, pro-vice-chancellor for education and students, has seen enough of the detectors’ flaws to decide the university’s energy is better spent elsewhere.
Multiple universities have restricted or disabled AI detection tools including those from Turnitin, GPTZero and Copyleaks. The decision follows mounting evidence of unreliability. False positives run high. The software often flags the careful English of non-native speakers. A New York court ruled in January 2026 in favour of a student handed a 100 per cent AI-generated score by Turnitin, exposing both technical limits and procedural weaknesses.
Institutions from Vanderbilt to Yale, Johns Hopkins, Northwestern, Waterloo, Cape Town and Curtin have already moved away from the tools. In the UK the shift is quieter but unmistakable. An audit of 163 institutions found more than 40 per cent have no publicly accessible AI policy. A separate study of 96 degree-awarding bodies put the figure at 41 per cent. Without clear rules, suspicion fills the gap.
Student behaviour and the fear of false flags
A survey of more than 6,600 students across seven UK universities revealed that 32 per cent admitted some level of unpermitted AI use. Yet another poll of 1,054 undergraduates, carried out in December 2025 by the Higher Education Policy Institute, showed 42 per cent were less likely to touch the technology at all for fear of being wrongly accused. The numbers sit uneasily together. Near-universal experimentation collides with widespread anxiety.
Rising numbers of AI-related misconduct cases have reached the Office of the Independent Adjudicator for Higher Education. Each complaint carries the same uncomfortable question: can the machine be trusted to separate honest work from shortcut?
These tools are still developing. False positives are high. AI-generated text can be modified and therefore unreliable. The focus should be on good assessment design and determining what is being assessed rather than stopping AI use.
Judy Williams’s words, reported by the Financial Times, capture the growing scepticism. Sam Illingworth, professor of critical AI literacy at Edinburgh Napier University, goes further. He argues that detection is not the solution. Many universities are now redesigning assessments to focus on oral components and the student process itself.