Generative artificial intelligence has moved from a specialist concern to an everyday feature of university life, putting pressure on institutions to redesign assessment and clarify what a degree demonstrates in an increasingly automated labour market.
A 2025 survey of 1,041 full-time undergraduates by the Higher Education Policy Institute and Kortext found that 92% had used artificial intelligence in some form, up from 66% the previous year. The proportion using generative AI for assessments rose from 53% to 88%. The survey report also said 59% believed their assessment methods had changed “a lot” in response to the technology.
The figures do not establish that students are routinely submitting machine-written work as their own. AI use can include brainstorming, summarising, translation, proofreading and research assistance, as well as producing text. But the scale of adoption has made older assumptions about take-home essays harder to sustain.
From prohibition to assessment redesign
Universities have responded unevenly. Some continue to prohibit unauthorised AI use, while others permit it under stated conditions and ask students to disclose how tools were used. The HEPI survey found that 80% of students agreed their institution had a clear AI policy, and 76% believed it would detect AI use in assessed work.
Those findings point to a confidence gap. Detection systems can generate false positives, particularly for students writing in a second language, while polished AI-generated prose can be difficult to distinguish from conventional editing. A policy based mainly on surveillance may therefore risk punishing legitimate assistance without restoring confidence in the assessment itself.
Instead, many educators are shifting attention towards assessments that are harder to outsource: oral examinations, in-class writing, practical projects, staged submissions and discussions about a student’s research process. Such approaches can offer stronger evidence of understanding, but they also require more staff time and may create accessibility or anxiety concerns if introduced without support.
“Institutions [should] keep their assessment practices under constant review,” the HEPI report concluded, arguing that staff need support to improve their AI literacy.
Degrees and the changing entry-level job
The debate matters beyond academic integrity. Employers are also reassessing which tasks should be assigned to new graduates as generative tools become capable of producing first drafts, basic analysis, code and routine communications.
Recent reporting on the 2026 graduate market, citing figures attributed to the New York Federal Reserve and the National Association of Colleges and Employers, described elevated underemployment among younger graduates and a growing emphasis on skills-based hiring. The same account reported that internship experience ranked highly among employers’ hiring considerations. The analysis should be treated cautiously because it brings together several datasets and projections, but its central question is widely shared: whether degrees still provide enough evidence of job readiness on their own.
Employers may value graduates who can use AI effectively while checking its errors, protecting confidential information and explaining decisions. These capabilities are difficult to capture through a single exam or credential. They also risk becoming advantages concentrated among students who have better access to paid tools, professional networks and relevant work experience.
Opportunity—and a new layer of inequality
Students can benefit from AI as a low-cost tutor, language aid or accessibility tool. It can provide immediate explanations and help users practise outside office hours. For students balancing employment, caring responsibilities or disability, that flexibility may be meaningful.
Yet unequal access remains a concern. Paid versions may offer stronger models and higher limits, and students differ in digital confidence and in the guidance they receive about responsible use. If universities assume that every student can independently master AI, existing inequalities may become less visible rather than disappear.
The emerging settlement is therefore unlikely to be a simple choice between banning AI and allowing it everywhere. Universities will need transparent rules, assessment formats that test genuine learning, staff training and support for students who cannot afford premium tools. Employers, meanwhile, will have to decide whether entry-level roles are designed to develop graduates—or merely to filter for experience they have not yet had the chance to acquire.
The immediate challenge is not whether AI belongs on campus. It is ensuring that its presence strengthens the value of learning instead of weakening trust in degrees or narrowing the path into work.