Back to blog

What breaks when you police student essays with AI detectors?

Thắng Đoàn
Thắng Đoàn

Imagine your strongest student accused of cheating because a detector flagged an essay they wrote themselves. That is happening, and regulators now admit the take-home essay no longer proves much at all.

Detectors punish honest students

Stanford research found that students who wrote essays without any AI were still flagged as ChatGPT users by detection tools, as reported by the Times of India. A detector that errors this way is worse than no detector. The student who did the work gets accused, while the student who edited AI output carefully walks through.

  • Stanford: detector false positives hit students who never used AI
  • Accusations fall hardest on students with the least recourse

Regulators admit coursework is broken

England's Office for Students says AI has made it close to impossible to assess coursework, per the Guardian. Its recommendation is blunt: universities should return to traditional exams and in-person presentations. The OfS also noted many institutions had already redesigned how they assess students. That is a regulator conceding the take-home essay no longer proves anything.

  • OfS urges in-person exams and presentations
  • Many universities had already redesigned assessment, the OfS noted

Schools are setting usage rules instead

New York City schools introduced a new policy on student AI access, and CNBC reports experts see promise in this approach for parents weighing kids' AI use. The shift treats AI as a skill to govern, not a crime to detect. Times Higher Education makes the same case for universities: when AI can produce a first draft, assessment should measure what the student does beyond that draft.

  • NYC policy shifts focus from detection to access rules
  • Assess the student's process, not just the polished text

What this does not solve

Returning to exams does not teach students to write well with AI, and it does not fix detector bias. None of the sources show a reliable way to prove who wrote a take-home text. Institutions that keep coursework must accept ambiguity or redesign tasks so the writing happens where it can be observed.

  • In-person exams bring back their own old problems
  • Governing AI use assumes teachers have time to coach it
Which of your assessments still prove the student did the thinking, if AI wrote the first draft?
If a detector falsely accused your strongest student tomorrow, what appeal process would they face?

List your current assessments and mark each one: can it be completed by AI alone, yes or no.

Sources

Times of India: Stanford research on AI detector false positives

Times Higher Education: Assessing writing when AI drafts

CNBC: NYC schools AI policy

The Guardian: OfS says AI makes coursework hard to assess

Share:

Recommended for you

Enjoyed this article?

Subscribe for new articles. No spam. Unsubscribe anytime.

By subscribing you agree to receive the newsletter. No spam, and you can unsubscribe anytime.