AI in education: unis are testing the wrong skill
94% of students use AI for assessed work. Exams aren't the fix

94% of UK undergrads now use AI to help with assessed work. And the big policy answer is... more exam halls.
HEPI's 2026 survey of 1,054 students found 95% use AI in some form.
Earlier this month, Policy Exchange told universities to end unsupervised remote exams "immediately".
Honestly that misses the point.
The debate about AI in education is all about catching cheats. The real problem is WHAT we've been testing.
How students are actually using AI at university
From the HEPI and Kortext report:
- 94% use generative AI to help with assessed work
- The share putting AI-generated text straight into assessed work has risen to 12%
- 65% say assessment has changed significantly in response to AI
- Only 38% say their uni provides AI tools, and just 48% feel teaching staff are helping them build these skills for their careers
So nearly everyone uses it, and fewer than half feel taught how to use it well.
That's the gap. Students aren't waiting for permission, they're just not being taught the judgement part.
The consensus fix: back to the exam hall
The Policy Exchange report wants:
→ unsupervised remote exams ended "immediately" → a return to in-person assessment → more detailed degree classifications that show what a graduate has actually learnt
It found 78% of the 120 universities it sent FOI requests to rely on online remote exams for summative assessment, and 70% plan to keep using unsupervised ones.
Fair enough on integrity. If you want to know a student did the work, watch them do it.
But here's the contrarian bit.
Why AI in education exposes the wrong skill
Most school and uni assessment rewards one thing: recalling the right answer under pressure.
AI made that skill cheap.
Locking the exam hall door protects the test. It doesn't protect the value of what's being tested.
What school and uni rarely teach, start to finish:
- turning a vague brief into a clear question
- deciding what a good answer looks like before you see one
- checking a confident answer against real sources
- showing which parts of the work are actually yours
What employers say they want is different.
The ISE Student Development Survey 2026, with 144 employers, lists critical thinking, AI literacy, communication and adaptability as rising in importance.
Routine admin, basic data and writing tasks are falling. And 87% expect AI to reshape graduate roles.
bad exam question: "list the causes of X" good exam question: "here's an AI's answer on the causes of X. What's wrong with it, and what should it have been asked instead?"
Both can be invigilated. Only one tests a skill that's still scarce.
(and two-thirds of those employers already think graduates use AI to misrepresent their skills... so faking it won't get you far either)
A critical-thinking framework: frame it, then judge it
Five steps for any assignment, with AI or without:
- Frame the problem. Rewrite the brief in one sentence. What does "done" look like? If you can't say, no prompt will save you
- Predict before you ask. Write down roughly what you expect the answer to be. Now you've got something to compare the output against
- Interrogate the output. Pick the three claims doing the most work and check each one against a real source
- Name your contribution. What's in this that the tool couldn't have produced? If the answer is nothing, that's a grade risk and a career risk
- Defend it out loud. Could you explain it in a five-minute viva? If the exam hall crowd wins, you might have to anyway
What to do this month
- Before term starts, take one past essay question and run steps 1 to 5 on it with a chatbot. Keep your notes
- Ask each module leader exactly how AI is allowed on each assessment. With 65% of students saying assessment has changed, don't assume last year's rules
- Practise explaining one topic out loud in two minutes, no notes
- Build one small thing outside your course where you had to frame the problem yourself
so TLDR: exam halls can stop the cheating, but they can't make memorising valuable again. Framing problems and judging answers is the skill.
What's one assessment you've done that actually tested your judgement?
Sources
- New HEPI and Kortext Report: Student Generative Artificial Intelligence Survey 2026Higher Education Policy Institute · 12 MAR 2026
- Student Generative Artificial Intelligence Survey 2026Higher Education Policy Institute · 12 MAR 2026
- Ban all remote unsupervised tests 'immediately', urges reportTimes Higher Education (reporting Policy Exchange) · 18 AUG 2026
- Entry-level work reshaped not replacedInstitute of Student Employers · 07 MAY 2026
Get the next issue by email
theVANTAGEd is written by students to bring critical thinking to AI across real industries. One issue every two weeks.