Generative artificial intelligence is forcing universities to reconsider one of higher education’s basic promises: that a submitted assignment shows what a student can do independently. As AI tools become routine in study and increasingly relevant in the workplace, institutions are shifting from trying to police every use of the technology toward redesigning assessment itself.
A 2025 survey by the Higher Education Policy Institute found that 92% of students had used AI in some form, up from 66% the previous year. The survey also found that 88% had used generative AI for assessments, compared with 53% in 2024. At the same time, 80% said their institution had a clear AI policy, suggesting that universities are responding, though not necessarily in a consistent way.HEPI survey
From prohibition to disclosure
The central challenge is distinguishing assistance from substitution. Students may use AI to brainstorm, explain difficult material or improve a draft, while submitting machine-generated work as their own can undermine the purpose of an assessment. Yet a simple ban can leave students uncertain about where the boundary lies.
Research involving students and staff at an Australian university found that students developed “individually unique and often complex ethical frameworks” around acceptable AI use. The researchers argued that policies should address enforcement, authentic learning and the emotional wellbeing of both students and teachers, rather than relying only on blanket permissions or prohibitions.Australian assessment study
Some universities have begun classifying assessments according to how AI may be used. Guidance described by the Higher Education Policy Institute includes categories such as AI-integrated, AI-supported and AI-prohibited work. The approach aims to make expectations explicit while acknowledging that graduates will encounter AI outside the classroom.Higher Education Policy Institute analysis
“Assessment practices” need “constant review” as AI tools become more powerful and students become more proficient with them, the HEPI survey report concluded.
Why assessment is changing
Traditional essays and take-home assignments remain valuable, but they are less reliable as stand-alone evidence of independent writing. Universities are therefore experimenting with oral examinations, supervised work, process journals, classroom writing and tasks that require students to explain and defend their decisions.
One proposal highlighted by University College London is the use of sampled viva voce examinations, in which a small proportion of students would explain their submitted work orally. Such assessments could preserve written assignments while testing whether students understand the arguments and evidence they presented.University College London
There is no consensus that AI assessment tools can replace academic judgment. A 2025 comparative study found that AI could provide broad contextual feedback, while human evaluators were more precise and better aligned with grading rubrics. Its authors recommended a hybrid approach rather than fully automated evaluation.Utrecht University research
Preparing graduates for an AI-shaped workplace
The debate is also connected to employment. Employers increasingly expect early-career workers to understand how to use AI, assess its output and apply professional judgment. The National Association of Colleges and Employers’ 2026 Job Outlook update said employers planned to increase hiring from the graduating class by 5.6%, but described growth as uneven and reported that 45% viewed the overall market as “fair.”NACE Job Outlook
That creates a difficult balance for universities. Graduates who never learn to work with AI may be disadvantaged, but graduates who rely on it without verifying information may lack the analytical skills employers need. Assessments that require students to critique AI-generated claims, document their process and demonstrate subject knowledge could test both capabilities at once.
Students, however, face unequal conditions. Access to paid tools, reliable devices and guidance varies, while unclear rules can make cautious students fear disciplinary action and encourage others to conceal their use. Universities will need to provide training and consistent standards if AI literacy is to become an educational benefit rather than another source of inequality.
The likely outcome is not the disappearance of essays or exams, but a broader definition of what counts as evidence of learning. Written work will increasingly be paired with conversation, reflection, practical application and verification. The institutions that adapt most effectively will be those that treat AI not merely as a threat to academic integrity, but as a reason to make learning more visible.