Do unproctored online assessments still work in the age of AI?
TL;DR: For measuring unaided human ability — no. AI assistants answer standard assessment questions instantly, so an unproctored score now reflects the candidate's tooling, not their knowledge. Unproctored formats still work for low-stakes practice and open-AI tasks. For anything where the score carries consequences, proctoring is what makes the number mean something.
The claim: AI broke the honesty assumption unproctored testing was built on
Evidence: Unproctored assessment always relied on a bet: most people won't cheat, and cheating takes effort. Generative AI deleted the effort half of that bet — a screenshot or a pasted question returns an answer in seconds, in another tab, with zero preparation. When cheating is effortless and invisible, honest test takers are the ones disadvantaged, and score distributions stop distinguishing ability. The problem isn't that everyone cheats; it's that you can no longer tell who did.
Where each format still works
| Assessment purpose | Unproctored OK? | Why |
|---|---|---|
| Practice quizzes, formative checks | Yes | Low stakes; learning is the goal |
| Open-book, AI-allowed tasks | Yes | AI use is part of the measured skill |
| Graded course exams | No | Grade integrity requires knowing who answered |
| Certification exams | No | The credential's market value depends on it |
| Pre-employment screens | No | Hiring decisions ride on the score |
| Live interviews and vivas | Needs monitoring | Coaching and copilots operate in real time |
Step-by-step: deciding what to proctor
- Ask what the score is for. If a wrong score costs nothing, don't proctor. If it feeds a grade, credential, or job decision, proctor.
- Decide the AI policy per assessment. "AI allowed and expected" is a legitimate design — but it must be a choice, not a leak.
- Match proctoring depth to stakes. Quizzes: lockdown and tab-switch detection. Finals and certifications: full identity, environment, and behavior layers.
- Keep friction proportional. Zero-install browser proctoring (paste a URL, students open a link) makes proctoring cheap enough to use exactly where it's needed.
- Communicate the rules. Students and candidates behave differently when monitoring is disclosed — deterrence does half the work before the exam starts.
- Review with evidence. Trust scores triage; humans decide from timelines and snapshots.
FAQ
Can't I just write AI-proof questions instead? Harder questions raise the bar briefly; frontier models keep catching up. Question design helps, but it cannot verify who answered — only monitoring does that.
Do AI-writing detectors solve this for essays? No. Their false-positive and false-negative rates make them contested as sole evidence. A monitored writing session proves authorship directly.
Is it fair to proctor some assessments and not others? Yes — that's proportionality. Stakes determine controls, and disclosing the policy up front is what fairness requires.
What's the minimum viable proctoring for a mid-stakes quiz? Browser lockdown with tab-switch detection, no identity layer. With Neuroxa.ai that's a pasted URL and a couple of toggles.
By Pinal Dave · Last updated: 2026-07-23