Discussion

AI exam cheating is turning test security into an arms race

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#14228

Exam authorities are tightening security against AI-assisted cheating, but a reported $499 tool that bypassed US SAT lockdown software shows how quickly those defences can be tested. The emerging contest matters not only for how exams are policed, but for whether assessment design can keep up.

Watch Desk analysis

What happened

The Korea Times, in an article translated from its sister publication Hankook Ilbo and edited by its newsroom, describes measures ranging from locked-down computers and bans on smart glasses to AI-powered surveillance systems. In the US, the College Board’s Bluebook app blocks access to other programmes during the SAT. The report says the board discovered AutoSAT, a $499 programme that could bypass Bluebook, send on-screen questions to an AI chatbot and display answers. At least four students used it in March, the report says. The College Board then modified its software and sued the developer. The article says the developers later released an updated version.

China has used AI-powered systems during its national university entrance exam to flag unusual behaviour and produce video clips for supervisors to review. The report also describes facial recognition and monitoring of keyboard and mouse activity at some overseas testing centres, alongside phone detectors, signal jammers and restrictions on wearable devices.

Why it matters

The examples show that AI is part of both sides of the contest: students can use it to answer questions, while authorities use it to monitor exams. Locking down a computer may raise the cost of cheating, but it is not a complete answer when a tool can be adapted to get around the lock.

The report also gives a sense of scale. It cites Ofqual figures recording 2,225 cases of cheating involving mobile phones or smart devices in UK secondary-school exams last summer, 44.3% of all malpractice cases. Those figures cover device-related cheating, not AI use specifically.

Our read

More monitoring may help protect the credibility of an exam, but there is a point where the surveillance starts to become its own problem. The report cites recommendations from an MIT committee to combine assessment formats, including supervised in-person exams and oral tests, rather than rely on AI detection alone. That is a more durable answer than assuming every new lock will outlast the next workaround. It also gives schools a harder, more useful question: what is this assessment meant to prove?

What to watch

  • Whether exam authorities change assessment formats as well as tightening device controls.
  • Whether restrictions on facial recognition and activity monitoring are clear and proportionate.
  • How exam boards respond as tools designed to bypass safeguards are updated.

Discussion spark: Should exam authorities prioritise tighter device surveillance, or redesign assessments so that bypassing a lockdown app matters less?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.