A personal reflection on an outstanding episode of Isabelle Gonthier’s Tried and Tested podcast, my own research into AI-powered cheating systems, and why effective security should not come at the expense of candidate trust.
I'm a regular listener of Isabelle Gonthier's Tried and Tested podcast. It's consistently a source of interesting conversations about the assessment and testing industry. But every now and then, an episode really stands out.
That was certainly true of Episode 62, Inside the Entrust Fraud Lab: The Evolution of Identity Fraud in Testing. Isabelle visited the Entrust Fraud Lab to speak with Simon Horswell, an experienced fraud specialist, about how identity fraud is evolving and what it means for high-stakes assessment.
This was a different kind of episode. Rather than discussing threats in the abstract, they explored the techniques behind them in a working fraud lab. I'd particularly recommend watching the video, because the demonstrations add a dimension that audio alone can't convey.
What stayed with me, though, wasn't just the sophistication of the demonstrations. It was Simon's description of technological convergence: tools developing for different purposes, becoming easier to use and combining to open up entirely new routes for fraud.
“The bar to entry is so much lower that it means that the attack surface is much greater than it's ever been.”
— Simon Horswell, Tried and Tested (10:50)
That observation connected directly with some of my own recent research. It also made me wonder whether the assessment sector is still looking for threats in the places we've become accustomed to watching.
Simon describes the progression from physical disguises and other presentation attacks to real-time face swaps and virtual-camera injection. The difference matters. Rather than merely putting an image in front of a lens, an injection attack can introduce manipulated imagery into the digital verification process itself.
It's not difficult to see the implications for remote testing. An impersonator could attempt to appear as the registered candidate, while someone else actually takes the examination. That is a real threat, and the work of fraud labs to understand and anticipate it is valuable.
For a quick sense of how accessible the underlying technology has become, take a look at AKOOL's face-swap application or its live camera face-swap demo. Seeing a face swap is not the same as defeating an identity verification system, but it makes the direction of travel easier to grasp.
My question goes a step further: what if the person sitting the examination really is the candidate, but they are receiving assistance that the security process cannot see?
“The capabilities of AI are significantly in the hands of the individual.”
— Tim Burnett, on the changing threat landscape
In that scenario, the identity check may be perfectly accurate. The candidate is genuine. Yet that alone tells us little about whether the resulting score reflects their unaided knowledge or competence.
Over recent months, I've been exploring what happens when multimodal AI, natural-language software development — often called vibe coding — and low-cost consumer hardware come together.
Software that once demanded specialist programming skills can increasingly be prototyped by describing an idea to an AI coding tool. Add readily available cameras, smart glasses and other consumer devices, and there are new possibilities for creating systems that operate away from the device on which the examination takes place.
This is not a claim that every inexpensive gadget can bypass proctoring or that AI-generated code always works. The significance is the falling barrier to experimenting: one motivated individual can test ideas that previously might have required a team, specialist equipment or substantial development time.
In my two-part research presentation, AI-Powered Cheat Systems, I investigate this convergence and demonstrate what I found, including work involving smart glasses and off-device approaches. The presentation is available by request through the Test Community Network; the page explains how to access it.
I summed up the particular concern this way in the research:
“The threat isn't on the device. Off-device threats are significantly more capable now, and all they need is line of sight of the screen.”
— Tim Burnett, AI-Powered Cheat Systems research
That is the parallel I drew while watching Isabelle and Simon. The same convergence that has made real-time impersonation easier to attempt may also make bespoke, AI-assisted cheating more accessible to individual candidates. One threat asks whether the person is genuine; the other asks whether the work is genuinely theirs.
One of Simon's analogies from the podcast deserves to stick:
“Fraud is like water. It will always find where the weak point is, where the leak is.”
— Simon Horswell, Tried and Tested (around 8:00)
It's a useful way to think about layered assessment security. If we invest heavily in checking faces, watching screens and enforcing browser restrictions, we still have to ask what happens beyond those controls. That is not an argument for abandoning them. It is an argument for understanding their limits, testing the assumptions behind them and thinking about the whole assessment process.
Simon is equally clear that there is no single cure:
“Nothing is a silver bullet in this industry.”
— Simon Horswell, Tried and Tested (16:08)
But there was another part of the conversation I found just as important: the candidate experience.
I keep coming back to airport security as an analogy. As passengers, we expect serious checks. We want to feel that the system is looking after everyone, not merely waving people through. A completely unprotected airport would hardly inspire confidence.
Yet we also appreciate technological improvements that make the experience smoother. On a recent journey, biometric recognition meant I barely needed to show my passport. Changes to baggage screening can reduce some of the old rituals of unpacking liquids and electronics, depending on the airport. Security can become less intrusive without necessarily becoming less effective.
Isabelle made a similar point about test takers during the episode:
“It matters to them as well because they want to make sure that someone else doesn't have an unfair advantage.”
— Isabelle Gonthier, Tried and Tested (19:13)
I agree. Candidates have a stake in the credibility of the qualification. They want to know there are genuine controls against unfair advantage. A reasonable amount of visible rigour can be reassuring.
But I also seem to attract extra screening involving my shoes with remarkable regularity. I don't know whether that reflects a particular trigger, a random check or simply a coincidence. What I do know is that, after enough unexplained interventions, the process starts to feel arbitrary.
“Where people start to lose trust is the number of false positives.”
— Tim Burnett, on candidate trust
To be clear, I can't know whether my own airport checks were false positives. But the perception of arbitrary checking is instructive. In assessment, a genuine false positive can be far more serious than a minor inconvenience: an innocent candidate may face interruption, distress or suspicion of misconduct.
This is where the phrase security theatre becomes interesting. We normally use it to describe checks that appear reassuring but don't deliver meaningful protection. Yet visible security can also serve a legitimate purpose: it tells honest candidates that the integrity of their achievement matters.
The point is not to add friction for show. It is to ensure that the checks people experience are necessary, proportionate and effective. Where protection can work discreetly in the background, so much the better.
Simon explains that technological progress is enabling precisely this shift: more ways to identify injection attacks without making candidates jump through extra hoops. That offers a useful model for the wider assessment sector.
“There's a need for some friction.”
— Tim Burnett, on purposeful friction
But not every interruption earns that justification. Repeated false alarms, opaque decisions and poorly handled alerts can undermine the trust they are meant to create. And a visible, reassuring process is not enough if newer off-device threats fall outside its coverage.
The goal, as I see it, is not zero friction at any cost. It is security in which every moment of friction earns its place.
My takeaway from the Fraud Lab episode is that assessment security cannot be treated as a fixed collection of checks. The technology has moved on; the barriers to building new tools have changed; and so have the ways people may try to circumvent assessments.
We need layered controls and better intelligence about emerging threats. We need to ask suppliers difficult questions about what their systems can actually detect. And we need to protect candidate confidence by minimising unnecessary interruption, handling alerts fairly and remembering that identity verification is only one part of assessment integrity.
If this resonates, I'd encourage you to explore the two pieces that prompted and informed this reflection:
Watch Isabelle Gonthier and Simon Horswell inside the Entrust Fraud Lab — Tried and Tested, Episode 62. The video is particularly worthwhile for the demonstrations and Simon's insights on evolving fraud techniques.
Explore my AI-Powered Cheat Systems research presentation at the Test Community Network. It examines the convergence of AI-assisted coding and affordable hardware, with a two-part presentation available by request.
The future of assessment security isn't simply about watching candidates more closely. It's about understanding the threats more intelligently, protecting the whole assessment process and making sure that the security candidates experience genuinely deserves their trust.