Generative artificial intelligence has moved from an experimental classroom tool to an everyday part of university study, creating a challenge that reaches beyond academic misconduct: what should a degree prove when students can ask software to explain concepts, draft essays and generate code?
A 2025 survey of 1,041 full-time undergraduates in the United Kingdom found that 92% had used generative AI in some form, up from 66% the previous year. The share reporting that they had used it for assessments rose from 53% to 88%, according to the Higher Education Policy Institute.
The figures do not establish that students are routinely submitting unedited machine-written work. They do show that the old boundary between independent study and automated assistance is becoming difficult to maintain. Students may use AI to brainstorm, translate, check grammar, summarise reading or produce an initial answer before rewriting it themselves.
Assessment is changing faster than policy
Universities have responded unevenly. Some have restricted AI use in particular assignments or returned to invigilated examinations. Others have incorporated the tools into coursework, asking students to critique generated answers, document their prompts or demonstrate how they verified sources.
The HEPI survey found that 59% of respondents believed their assessment had changed “a lot” because of generative AI. Students also expressed limited enthusiasm for the prospect of exams assessed by AI: 34% said such a system would make them work harder, while 29% said it would make them work less hard and 27% expected no change.
That ambivalence reflects a central problem. Automated detection systems can produce false positives, while unrestricted take-home essays may no longer reveal what a student can do unaided. Yet replacing written work with high-pressure exams can disadvantage students with disabilities, caring responsibilities or limited access to quiet study space.
“The question is not whether students will use AI, but whether universities can design assessment that makes learning visible.”
The quote above is an analytical framing rather than a statement attributed to a particular researcher. The evidence points toward a broader redesign: oral explanations, staged projects, practical demonstrations and reflective accounts may make it easier to distinguish genuine understanding from outsourced production, but they also require more staff time and consistent standards.
A workplace skill—or a shortcut?
Employers are simultaneously signalling that graduates will need to work with AI. A 2025 summary of Digital Education Council research reported that 86% of students used AI in their studies, while 48% said they did not feel adequately prepared for an AI-enabled workplace. That gap suggests that universities face pressure to teach both technical fluency and the judgement needed to question automated outputs.
Supporters of classroom AI argue that banning it leaves students unprepared for professional life. In many workplaces, employees will be expected to use generative tools for research, drafting, data analysis or customer support. Learning when to use them, how to protect confidential information and how to check errors could be as valuable as learning the tools themselves.
Critics counter that convenience can become dependency. If students use AI to avoid struggling through difficult material, they may finish a course with polished submissions but weaker writing, reasoning and subject knowledge. The risk is especially serious in fields where mistakes carry public consequences, including medicine, law, engineering and education.
The inequality question
Access is another fault line. Paid services may offer more capable models, higher usage limits and better file or image handling than free versions. Students with reliable devices, strong digital skills and supportive institutions may therefore gain more from AI than classmates who lack those resources.
Universities also face a transparency problem. Rules that differ by course can leave students unsure whether using a tool for grammar correction is acceptable, or whether asking it to generate an outline crosses the line. Clear, assessment-specific guidance is more useful than blanket warnings, particularly when policies explain how legitimate assistance should be acknowledged.
The immediate future is unlikely to be a return to an AI-free campus. The more consequential choice is whether universities treat generative AI as a policing problem or as a reason to clarify the knowledge, skills and judgement their degrees are meant to certify. As the technology becomes routine, the credibility of a qualification may depend less on preventing every use of AI than on proving what students can understand, explain and responsibly apply.