Universities are rewriting how they assess students as generative artificial intelligence becomes part of the workplace—and as institutions confront a basic question: what should a degree prove when software can draft, summarize and code in seconds?

The emerging answer is not a single policy. Across recent university guidance, institutions are moving toward assessment systems that distinguish between responsible AI assistance and the outsourcing of a student’s thinking. Assignments may permit tools for brainstorming or feedback, while other tasks restrict or prohibit them because they are intended to measure individual knowledge, judgment or performance.

From prohibition to disclosure

The University of Greenwich’s 2026/27 guidance divides assessments into categories, including work where AI is restricted, allowed for defined purposes or deliberately incorporated into the task. Students remain responsible for the accuracy, evidence and originality of what they submit, and are expected to explain how tools contributed to their work. [1]

That approach reflects a broader shift away from blanket bans. The University of Salford’s policy says assessment guidance should generally allow appropriate uses of generative AI, while prohibitions must be clearly justified and explained in advance. It also says students should be supported to meet learning outcomes with or without AI. [2]

For students, clearer rules could reduce the uncertainty that followed the rapid arrival of tools such as ChatGPT. Yet disclosure requirements create new work: students may need to record prompts, explain revisions or submit reflections alongside an essay. The University of Glasgow’s guidance for 2026/27, for example, allows course coordinators to set specific limits while requiring assessment instructions to state which AI-use scenario applies. [3]

Testing what students can do alone

The policy debate is also changing the shape of assessment. Universities increasingly face pressure to verify that students can reason and communicate independently rather than merely produce polished text. Possible responses include supervised examinations, oral defences, practical demonstrations and staged assignments that show how an argument developed.

These methods have advantages but also costs. Oral examinations and invigilated tasks can offer stronger evidence of individual ability, yet they require more staff time and may disadvantage students affected by anxiety, disability or unequal access to quiet study environments. A balanced system therefore needs both authentic AI-enabled work and accessible ways to demonstrate unaided competence.

“Students remain responsible for everything they submit,” the University of Greenwich says, including the accuracy of information, quality of evidence and originality of their work. [1]

The principle matters beyond academic integrity. AI systems can produce plausible but false claims, fabricated citations and biased conclusions. Students who use them without verification may complete an assignment faster while learning less—or carrying unreliable habits into professional settings.

A workplace skill, not just a study aid

Employers are unlikely to treat AI literacy as a specialist concern limited to technology degrees. A 2025 study published in Frontiers in Artificial Intelligence reported that graduates with stronger AI skills were more likely to be employed in work related to their field and to perceive gains in productivity and professional alignment. The study also associated basic AI knowledge with higher unemployment or employment outside graduates’ areas of training, although its findings do not establish that AI ability alone caused those outcomes. [4]

That evidence supports teaching students how to question outputs, protect confidential information, cite sources and decide when not to use a tool. It does not mean every discipline should adopt the same rules. A nursing assessment, a law problem, a programming project and a visual-arts portfolio may require different forms of AI use and different safeguards.

The central risk is unequal access. Students with better devices, paid subscriptions or more guidance may gain advantages that are invisible in final submissions. Universities that permit AI therefore need to provide alternatives and instruction, rather than assuming that all students can experiment on equal terms.

For now, the direction is clear: the degree is being asked to certify two abilities at once—the capacity to work effectively with new tools and the capacity to think when those tools are unavailable, unsuitable or wrong. How universities balance those claims will shape not only assessment, but the credibility of graduate qualifications in an AI-shaped labour market.

Sources