The Fear Was Already There Before Anyone Built a Detector

In December 2024, Savanta interviewed 1,041 full-time undergraduate students online, margin of error about 3%, on behalf of the Higher Education Policy Institute and Kortext. The result was Policy Note 61, Student Generative AI Survey 2025, written by HEPI's Josh Freeman with a foreword from Professor Janice Kay CBE, published on 26 February 2025.

The headline number: 92% of students now use some AI tool. Last year it was 66%. For assessments specifically, 53% of students used generative AI like ChatGPT last year. This year it's 88%. The share who say they've steered clear of AI for an assessment fell from 47% to 12% in twelve months.

Freeman's own line on this: "It is almost unheard of to see changes in behaviour as large as this in just 12 months." He's the Policy Manager at HEPI, and the report calls the increase unprecedented compared with the previous year's survey. Both are restrained descriptions of a number that moved 26 points in a single academic cycle.

The interesting part is what students said scares them.

What's actually keeping them up

Two things put students off using AI, according to the report. Being accused of cheating, at 53%. Getting false results or "hallucinations," at 51%. Everything else trails badly — the environmental cost of AI tools worries just 15% of respondents, which tells you where the anxiety actually sits.

Here's a student, quoted in the report:

"I enjoy working with AI as it makes life easier when doing assignments however I do get scared I'll get caught."

Here's another, describing the institutional guidance they're working under:

"They dance around the subject. It's not banned but not advised, it's academic misconduct if you use it but lecturers tell us they use it. Very mixed messages."

Read that second quote twice. That student is describing a policy that contradicts itself in real time — a rulebook where the rule and the exception are the same sentence.

The mechanism: nobody agreed on the rules, so everyone's guessing

Here's the part that explains the fear without needing a villain. It's a rational response to inconsistent signals coming from the institution itself.

80% of students say their institution's AI policy is "clear." 76% think their institution would spot AI use in an assessment. Read those two numbers next to the "mixed messages" quote above and something doesn't add up. A policy can be clear on paper and still leave a student staring at a submit button at 2am with no idea whether the thing they just did counts as help or as cheating.

The staff side is shakier. Only 29% of students agree their institution actively "encourages" AI use, and 40% disagree outright. Meanwhile the number of students who think staff are "well-equipped" to work with AI jumped from 18% in 2024 to 42% in 2025. That's real progress. It's also still under half. Two-thirds of students, 67%, think using AI is "essential" in today's world. Only 36% have actually received training in AI skills from their institution. Students are told the skill is essential and left mostly to teach themselves.

That gap — essential but untaught — is where the fear lives. You can't be trained out of a mistake you were never shown how to avoid.

The effort question

Assessment itself is shifting under this. 59% of students say the way they're assessed has changed "a lot" because of generative AI. The report asked what happens to effort if exams start being marked partly or mainly by AI. A third of students, 34%, said they'd work harder. Another 29% said they'd put in less effort. The remaining 27% said it wouldn't change anything for them.

Split roughly into thirds, then. One part pulls one way. One part pushes the other. A chunk of the cohort shrugs. That's what happens when you change the grading mechanism without telling students what it now rewards.

The gap runs along old lines

The report flags something uglier underneath the averages: a widening divide in who actually uses these tools. Male students use AI more than female students. STEM and Health students use it more than other disciplines. Socioeconomically advantaged students use it more than their peers. The tool that's supposed to level the playing field tracks pretty closely with who already had an advantage before it existed.

That matters for the fear numbers too. If AI use skews toward students with more resources and more confidence navigating institutional ambiguity, the students most exposed to an accusation of misconduct may be the ones with the least cushion to survive one.

What HEPI told universities to do about it

The report's recommendations are simple. Stress-test assessments — check, urgently, whether they can be completed by AI without anyone noticing, before a plagiarism panel finds out for you. Adopt a policy that supports AI skills while being honest about the risks. Retire the "banned but not advised" doublespeak one student described. Collaborate across the sector on best practice, with a body like Universities UK positioned to lead it, so every institution starts from something better than a blank, inconsistent page.

None of that fixes the fear this year. It's a roadmap for next year's survey to look different.

What this means for the 2am version

If you're a student reading this at the point in the term where everything is due at once, here's the honest read. AI use on assessments climbed for a reason — it saves time (51% said so) and it improves the quality of the work (50%), which is exactly why it's now normal.

The threat is the gap between what the tool produces and what you can actually stand behind when someone asks you to defend it. A source that doesn't exist is what turns "I used AI to help" into "I got accused of cheating." The fear students reported is about not knowing, until it's too late, whether what the technology handed them will survive someone checking it.

That's the boring, fixable part. Before you hand anything in, every source and claim you're relying on can be checked against the live record. The source exists. It says what you think it says. It's still standing, and a retraction since you found it gets flagged. It stays out of your decision to use AI and leaves your argument alone. It just closes the specific gap that turned 53% of a national student survey into people scared of their own coursework.

Boring is the goal. Boring is turning something in and never finding out the hard way what "hallucination" meant.