AI Proctoring Meets AI Cheating: When Technology Fails Everyone

The promise of remote testing was simple. Students could take an exam from home while artificial intelligence monitored for suspicious behavior, reducing the need for large testing centers and making education more accessible.

Instead, one of the largest AI-proctored exams in recent memory turned into a cautionary tale.

According to Ars Technica, approximately 58,000 students will have to retake an entrance examination after officials concluded that widespread AI-assisted cheating likely compromised the results.

What happened?

The university offered applicants the ability to take an entrance examination online. To protect the integrity of the test, several security measures were put in place, including:

• AI-powered monitoring through a webcam and microphone.
• Human proctors reviewing suspicious activity.
• A locked-down browser intended to prevent students from accessing other applications or websites during the exam.

On paper, this sounds like a comprehensive solution.

Unfortunately, the results suggested otherwise.

Officials noticed that unusually high scores appeared across the applicant pool. Students earning exceptionally high marks increased by nearly five times compared to previous years, raising immediate concerns that many examinees had found ways to use generative AI despite the monitoring systems. (Ars Technica)

The problem with AI policing AI

This story illustrates an emerging challenge.

Educational institutions are increasingly relying on AI to detect cheating.

Students, meanwhile, have increasingly capable AI tools that can answer questions, solve math problems, summarize text, write essays, and even explain complex subjects in seconds.

The result is an arms race.

Every improvement in AI detection is quickly met with new techniques for avoiding detection or using AI in ways that appear legitimate.

That makes it increasingly difficult to determine whether a student actually understands the material or simply knows how to work with an AI assistant.

False positives are also a concern

One aspect of these systems worries me just as much as cheating itself.

We’ve already seen AI systems produce false positives in other areas.

Apple’s recently discussed nudity detection issues demonstrated that machine-learning systems can occasionally flag perfectly innocent content as something it is not.

If AI can mistakenly identify harmless photos, could similar technology eventually misidentify perfectly honest students as cheaters?

That is a question educational institutions should take seriously.

An accusation of academic dishonesty can have significant consequences, and any automated system should provide a way for human reviewers to carefully evaluate questionable cases before penalties are imposed.

The honest students pay the price

Perhaps the biggest takeaway is who suffers the most.

It isn’t necessarily the students who cheated.

Instead, it is the honest applicants who prepared, followed the rules, and earned their scores fairly.

Because officials no longer trusted the results, tens of thousands of students now have to repeat the examination.

That means additional stress, more preparation, more scheduling, and weeks of uncertainty.

When a testing system loses credibility, everyone loses.

Will remote exams survive?

Remote testing is unlikely to disappear.

It provides accessibility for students who live far from testing centers, have disabilities, or face other barriers to traveling.

However, this incident demonstrates that simply adding AI monitoring is not a complete solution.

Educational institutions may need to reconsider how they evaluate knowledge by:

• Returning some high-stakes exams to in-person testing.
• Using oral examinations where students explain their reasoning.
• Designing questions that emphasize critical thinking over memorized answers.
• Combining multiple assessment methods instead of relying on a single online exam.

None of these approaches are perfect, but they may prove more resilient than relying on AI alone.

Final thoughts

Artificial intelligence is becoming a valuable educational tool. It can tutor students, explain difficult concepts, and help people learn more efficiently.

But when AI is used both to supervise exams and to help students answer them, institutions enter a technological arms race that may have no clear winner.

This case is a reminder that technology should support education, not replace sound assessment practices. As AI becomes more capable, schools, universities, and certification organizations will need to rethink not only how they detect cheating, but also how they measure genuine learning in the first place. (Ars Technica)