How a £7 Webcam, with a Free Vibe-Coded AI App, Could Be a Problem for Test Security

TLDR: Tim Burnett built a working exam-cheating system using a £7.36 USB webcam from Amazon and a free AI app created through natural language prompts — no coding required. It scored 8 out of 10 on a GCSE algebra test. Counter-fraud investigator Lee Fawcett, who is familiar with this kind of technology, is not surprised but offers advice on what assessment bodies should do next.

Watch the USB cam demo and the conversation with Lee in full here >

What does it take to cheat on an exam in 2026? Not a criminal network. Not specialist hardware. Not a shady contact in a back room. A £7.36 USB webcam from Amazon, a free app built through natural language prompts on Lovable, and about an afternoon of tinkering. That was enough to score 8 out of 10 on a GCSE algebra test, with almost no technical expertise involved.

That's the proof-of-concept Tim Burnett demonstrated on the latest episode of the Test Community Network podcast, before sitting down with Lee Fawcett, a counter-fraud and investigation specialist who works with malpractice teams across the UK qualifications and assessment industry, to talk about what it means.

"We're the people that you don't want to have to call. We investigate malpractice, and it's normally when you have an issue that's outside your sanction scope."
— Lee Fawcett, Counter-Fraud and Investigation Specialist

How does the cheating setup actually work?

The setup is disarmingly simple. A miniature USB webcam, barely more than a bare circuit board, is taped to the edge of a keyboard, angled to capture the exam screen. It feeds into a laptop running an AI-powered app that Tim built using Lovable, a vibe-coding platform where you describe what you want in plain English and the tool generates the code. The app captures the screen every ten seconds, sends the image to an AI model, and reads the answer back through audio. No programming knowledge required.

On a GCSE algebra practice test, the system returned the correct answer for 8 out of 10 questions. It stumbled on a couple, misreading a minus sign in one equation and misinterpreting a rearrangement in another, but those are prompt-engineering problems, not fundamental limitations. A bit more refinement and the accuracy would climb further.

"Am I surprised you can do that? No. Not at all, because I've built something similar myself. And it works really well."
— Lee Fawcett, Counter-Fraud and Investigation Specialist

Why is self-service AI cheating different from what came before?

Contract cheating, paying a third party to sit your exam or write your coursework, has been around for years. Lee has evidence of thousands of candidates using these services across the sectors he works in. But those routes carry risks for the cheat: criminal organisations might use it as leverage for blackmail, and even colluding with a friend creates a relationship that could turn sour. The power dynamic always sits with whoever helped you cheat.

Self-service AI cheating changes that dynamic entirely. There's no third party to pay, no one to blackmail you, and no trace of collusion. An individual with no technical background can build a working system alone, using free tools and cheap hardware readily available on Amazon.

"I don't think it's going to be long at all before someone commercialises this. Perhaps it will be in a Coke can or a coffee mug, and you'll be able to sit it on your desk and it will effectively feed you the answers."
— Lee Fawcett, Counter-Fraud and Investigation Specialist

Can online proctoring detect this kind of cheating?

Online proctoring was widely adopted during COVID and many organisations have continued with it. But Lee is frank about its limitations when it comes to concealed hardware and AI-assisted methods. The camera is tiny. The earpiece is invisible. The app runs on a separate device. None of the proctoring platforms he's tested have been able to detect the full range of tools in his sample kit.

That doesn't mean remote proctoring is useless, for high-volume, lower-risk assessments, it may still be the right delivery method. But the decision has to be grounded in a proper threat assessment, not in a general sense of comfort with the technology.

"If you're not in control of the environment, then you can't guarantee the security of the test. It's as simple as that."
— Lee Fawcett, Counter-Fraud and Investigation Specialist

What should assessment organisations do first?

Lee's advice is to start not with the technology, but with the qualification itself. What does the end credential represent? The value of what sits behind the assessment should determine the level of security applied to it.

From there, the approach Lee recommends is red teaming: stepping out of the exam-owner mindset and actively trying to break your own assessment. He's facilitated these sessions with malpractice teams and says the shift in perspective is immediate; teams already know their vulnerabilities, they just haven't been asked to think about them from the cheater's side.

"You need to think of it not as an exam owner. You need to think of it as a cheater. You need to red team this."
— Lee Fawcett, Counter-Fraud and Investigation Specialist

What to do next

The Conference on Test Security, which normally takes place in the United States, is coming to Cambridge in November. It's a niche but highly relevant event for anyone working in assessment security, and both Tim and Lee will be there. If you have a case study or presentation to submit, the deadline is imminent.

For organisations wondering where to start: assess the real-world value of your credential, red team your delivery method, and don't assume your proctoring platform has you covered. The barrier to cheating just got a lot lower.

Listen to the full episode: Test Community Network Podcast

Connect with Lee Fawcett on LinkedIn: linkedin.com/in/leefawcettuk

Connect with Tim Burnett on LinkedIn: linkedin.com/in/tburnett