Adapting to an AI-Driven World
This wasn’t just any panel. For this roundtable on cheating and AI in high-stakes assessment, we brought together voices from every corner of the exam experience – from awarding bodies and test centres to proctoring experts and product design. The goal? To have a frank, nuanced conversation about integrity, innovation and what comes next.
Robert Burns, Delivery Lead – Assessments, Institute of Chartered Accountants of Scotland (ICAS)
Robbie leads assessment delivery at the Institute of Chartered Accountants of Scotland. He brought a practical view from the awarding body side, with a strong focus on fairness, exam design and adapting to AI in a professional testing environment.
Owen Wood, Commercial Director, TEAMCo
Owen oversees high-stakes test centre operations across the UK. He brought real-world experience into candidate behaviour, invigilation challenges and how the physical testing environment influences integrity.
Justin Depuydt, Director of Strategic Operations and Partnerships, Proctorio
Justin shared the remote proctoring perspective, offering expertise on AI-driven monitoring, data privacy and how to balance security with a respectful candidate experience.
Jeroen Fransen, Head of Product, Cirrus Assessment
Jeroen represented the platform view, speaking about how integrity should be embedded into the design of assessments – from item creation to delivery and marking – not added as an afterthought.
Where Do We Draw the Line?
The roundtable kicked off with a big question: Where do we draw the line between support, preparation and outright cheating? With tools like ChatGPT, Grammarly and AI-powered devices becoming part of everyday life, the answer isn’t as clear-cut as it used to be.
Owen Wood from TEAMCo described how cheating has evolved far beyond notes hidden in socks. Today, invigilators are watching out for smart glasses, earpieces and even high-tech shirt buttons. “Once you think you’re on top of something,” he said, “something new gets introduced.” These days, invigilation means more than just rule enforcement, it requires experience, instinct and constant information-sharing.
But the real shift may be in perception. Justin Depuydt from Proctorio pointed out that many students genuinely don’t think they’re doing anything wrong. “There’s a disconnect,” he said, “especially with younger learners who grew up with these tools.” He suggested a simple rule: if a student uses something that helps them appear more skilled or knowledgeable than they actually are, that’s cheating, regardless of whether it’s old-school crib notes or new AI tools.
“If a student uses something that helps them appear more skilled or knowledgeable than they actually are, that’s cheating, regardless of whether it’s old-school crib notes or new AI tools.”
Justin Depuydt, Director of Strategic Operations and Partnerships, Proctorio
This ambiguity creates what Justin called the “soft middle”: students who wouldn’t normally cheat but go along with it because they think it’s normal. And that’s a hard problem to solve from a technology or proctoring perspective. It needs to start earlier, in the classroom, through open conversations and clear expectations.
Robert Burns from ICAS added that some universities are having students sign academic honesty pledges before they even begin their courses. It might sound strict, but it’s helping. “It’s about making sure students understand what cheating actually looks like, before they find themselves in a grey area,” he said.
It’s about making sure students understand what cheating actually looks like, before they find themselves in a grey area.
Robert Burns, Delivery Lead – Assessments, Institute of Chartered Accountants of Scotland (ICAS)
Still, the grey areas are growing. One story involved a student who rigged a phone into their desk and read questions out loud, not for themselves, but for the AI to hear and answer. The proctor eventually caught on, but only because they noticed the student repeatedly glancing at a specific spot. As Robert noted, “A less vigilant proctor might’ve missed it.”
In the end, the panel agreed: it’s not just about detecting cheating, but about understanding why students cheat. As Justin put it, some learners see passing the exam as the goal, not mastering the skill. “The certification becomes the obstacle,” he said, “not the achievement.” If we want to reduce misconduct, we need to rebuild a culture where learners are proud to show what they know, not just focused on getting through.
Have We Built a System That Rewards the Wrong Behaviours?
Once we started asking what counts as cheating, the conversation quickly turned to a deeper question: Why do people cheat in the first place? The panel suggested that the answer lies less in individual intent and more in the design of our assessment systems.
Jeremy Fransen observed that when students are trained to “pass the test” rather than understand the material, cheating becomes a logical response. “In the Netherlands,” he said, “we call it ‘teaching to the test.’ You’re preparing kids to pass, not to learn.” That distinction matters. Because if passing, not mastery, becomes the goal, then the method, cheating or not, starts to feel irrelevant.
Owen Wood added that AI complicates things even further. For students with learning differences or anxiety, AI tools can be a lifeline. But where’s the line between support and dependency? “It’s a fine balance,” he said. “AI can enable someone to show what they know, or it can give the illusion of competence they don’t actually have.”
It’s a fine balance. AI can enable someone to show what they know, or it can give the illusion of competence they don’t actually have.
Owen Wood, Commercial Director, TEAMCo
Robert Burns pushed the conversation further, calling out how outdated many assessments still are. “We’re still evaluating people like we did in the ’70s and ’80s,” he said, “but the world is different now. People don’t need to retain information in the same way, they can easily look it up. So are we really testing the right skills?”
Justin Depuydt offered a striking analogy: just as calculators replaced slide rules, maybe AI is forcing us to rethink what we actually need to assess. “We should be testing for the skills people need to do the job, not outdated tasks that no longer apply.”
Still, Jeremy raised an important caution: AI isn’t a calculator. “Overuse of AI can erode critical thinking,” he warned. “Employers need people who can think, especially when the AI output isn’t trustworthy.” It’s not about banning AI. It’s about teaching students how to question it, challenge it and use it responsibly.
Overuse of AI can erode critical thinking. Yet critical thinking becomes more essential than ever because employers need people who can question machines that aren’t necessarily trustworthy.
Jeroen Fransen, Head of Product, Cirrus Assessment
The group agreed: if we want to prevent misconduct, we need to fix what the system rewards. That means shifting from rote memorisation to authentic, relevant skills. It means teaching critical thinking, and making sure it’s something learners are expected to do, not something they’re left to figure out on their own.
And, as Jeremy put it bluntly: “It’s human nature to find the path of least resistance. If we design a system that rewards shortcuts, people will take them. The solution? Change what we reward.
Should AI Get a Seat at the Exam Table?
The panel explored a provocative question: should AI stay behind the scenes? Or take an active role in assessment itself?
Justin proposed the idea of a controlled AI assistant within the exam, one that could rephrase confusing questions or offer clarification without giving away answers. It’s not about giving students shortcuts, he argued, but about helping them demonstrate what they actually know, especially when flawed questions get in the way.
Others, however, raised concerns. Owen and Robbie highlighted the security risks: how do we prevent students from spoofing AI tools or bypassing restrictions? And how do we ensure AI outputs are consistent and fair, especially since large language models can behave unpredictably?
Still, the potential is hard to ignore. Used carefully, AI could improve accessibility, support neurodiverse learners and even reduce stress. “If it’s the right AI, at the right time, for the right reasons,” said Owen, “then yes – bring it on.”
The takeaway? AI might deserve a seat at the table, but only if we, as educators and assessment leaders, set the terms. Not tech companies. Us.
Can We Design Assessments That Don’t Need to Be Policed?
The final discussion looked to the future: What if we designed exams that made cheating irrelevant?
Jeremy kicked off with a simple but powerful idea: design exams that candidates want to take. When students are proud of what they’ve learned and excited to show it, the pressure to cheat fades. He also pointed to adaptive learning as a path forward: embedding assessment into the learning journey so subtly that students barely realise they’re being tested.
Design exams that candidates want to take. When students are proud of what they’ve learned and excited to show it, the pressure to cheat fades.
Jeroen Fransen, Head of Product, Cirrus Assessment
Robbie suggested that high-stakes, one-off exams might be giving way to broader portfolios of evidence, where critical thinking, ethics and real-world skills matter more than memorisation.
But scaling these models is a challenge. Justin Depuydt warned that personalised, engaging exams are great in theory, but tough to roll out at scale. The pressure to test large cohorts efficiently often drives institutions back toward rigid, easily policed formats.
Technology should support better design, not just enforce rules. Justin likened the role of proctoring to fencing a garden: not to trap people in, but to create a calm space where honest learners can thrive. “We’d love for legitimate test takers not even to notice we’re there,” he said, “but for those looking to cheat, we should feel present.”
Owen added that while security matters, so does empathy. Skilled human invigilators can ease stress, read the room and respond in real time. Something AI can’t replicate. “It’s about striking the right balance between support and surveillance,” he said.
And that was the final takeaway: exams of the future won’t be unmonitored, they’ll be well designed. Human-led, tech-supported and built to reflect the real skills learners need. Because when assessments are relevant, fair and thoughtfully constructed, they don’t need heavy policing. They speak for themselves.
Across the roundtable, one thing became clear: cheating isn’t just a problem of bad behaviour, it’s a symptom of deeper issues in how we design, deliver and value assessment. As technology evolves, so must our approach. That means rethinking what we test, how we test it and why.
Want to hear the rest? Register to get the full recording and discover what our panel thinks comes next.
If you have any questions or want to continue the conversation, feel free to reach out to us.
Want to dive deeper into the impact of ChatGPT and AI on assessment? Check out more of our insights here.
