How Exam Cheating Rings Automate Answers in Real Time
The second device broke the model
Remote testing removes the travel and scheduling burden of a test center, but it also removes the one thing a physical test center offered by default: a proctor who can see the whole room. A single human proctor watching a live video feed cannot reliably catch a hidden phone, a second screen, a coached answer, or a subtle handoff, and cannot review what happened after the fact unless the session was recorded.
Then ChatGPT shipped in November 2022 and one assumption broke: that the work in front of the student was the student. Generative AI on a second device became the default failure mode. The proctored laptop sits in front, the phone with ChatGPT sits under the desk, and eye-tracking catches a blatant look-down pattern but does not catch a student who memorizes the question, glances down once, and types. Multi-device cheating is the shape of the problem now, and the camera never sees it.
On-screen AI overlays made it worse. Cluely launched in April 2025, founded by an ex-Columbia student and backed by roughly $15 million from a16z, selling an on-screen assistant it markets as invisible to screen share and screen recording. When the cheat lives in the same monitor the proctor watches through, screen capture stops being evidence. The proctor sees a clean screen. The student sees the answer.
Why detection alone keeps losing
Automated cheating detection is not reliable enough to stand alone, and the evidence has piled up. Several institutions, including Yale, Vanderbilt, and Johns Hopkins, disabled Turnitin's AI writing detection feature, citing concerns about false positives. A 2023 study across seven detectors found a mean 61 percent false-positive rate on essays by non-native English writers, and OpenAI retired its own detector. A system that flags honest students at that rate cannot be the verdict.
The failure runs both ways. AI-only proctoring catches the obvious and false-flags everyone else. A live human proctor watching a handful of video feeds loses attention after the first thirty minutes, and every additional proctor adds cost. The architecture that holds up in 2026 is hybrid: AI surfaces the top few percent of moments to a trained reviewer, who judges with the full session recording available to pause, rewind, and examine in context.
Cross-session analysis is the part a single proctor can never do. Some systems, such as the Duolingo English Test, analyze patterns across thousands of test sessions at once, for example near-identical answers from test takers in the same region. A cheating ring that distributes answers to dozens of candidates is invisible one session at a time and obvious across a thousand. That is the detection that actually catches rings, not cameras.
What a defensible setup looks like
Verify identity before the exam starts and repeat through the session. Government ID checks, facial matching, and on some exams voice biometrics confirm the person taking the test is who registered, and re-checks catch a substitution mid-exam. Browser lockdown blocks tab switching, copy-paste, screen sharing, and unauthorized applications, which stops the lazy cheat even if it cannot stop the determined one.
Flag, do not decide. AI signals go to a trained human reviewer as a diagnosis to investigate, with the full session recording available, rather than triggering an automatic finding of misconduct. Treating every flag as proof of cheating is how institutions end up disciplining students whose only crime was looking away from the screen.
And check the regulation before the rollout. The EU AI Act classifies proctoring as high-risk, and the Digital Omnibus pushed those obligations to December 2, 2027. The SAT, Korea's CSAT, and China's gaokao have all added smart-glasses rules. A proctoring setup that ignores the legal frame is a setup that gets switched off the week before the exam.
Keep reading: Answer sharing on timed quizzes: the leak is usually a person, not a bot, How Bots Game Limited Releases, Quizzes and Gated Pages.
Frequently asked questions
What are the main cheating methods in 2026?
A second device running ChatGPT, an on-screen AI overlay like Cluely that markets itself as invisible to screen share and screen recording, AI glasses, and voice clones that need only a few seconds of audio. All of these post-date the camera-on-face vendors, and the vendors are patching, not rebuilding.
Why did remote proctoring break after 2022?
Because the work in front of the student is no longer necessarily the student's. ChatGPT shipped in November 2022 and broke the assumption that a watched screen means watched work. A student can memorize the question, glance down once at a phone, and type an answer the camera never sees.
What proctoring architecture actually works now?
Hybrid: AI surfaces the top few percent of flagged moments to a trained reviewer, who judges with the full session recording available. AI-only catches the obvious and false-flags everyone else, and live human proctoring does not scale past a few thousand sessions a month.
Does the EU AI Act regulate exam proctoring?
Annex III of the EU AI Act names student-monitoring AI outright as high-risk. The Digital Omnibus, in force since July 27, 2026 as Regulation (EU) 2026/1744, pushed high-risk obligations to December 2, 2027. Any proctoring vendor operating in the EU has to meet those obligations, not just claim to.