The Rise of ‘Invisible Cheating’: Why the UK Education System Needs to Rethink Online Exam Security

AI-generated answers are only part of the problem. UK institutions now face increasingly sophisticated forms of digital exam cheating.

The conversation around cheating in exams has been dominated by generative AI, but universities are at risk of solving the wrong problem. The bigger challenge emerging across the UK’s education system is that cheating has become increasingly invisible. Hidden browser overlays, second devices, wearable technology and organised answer-sharing networks are making misconduct harder to identify, exposing the limits of traditional approaches to exam security.

For decades, academic integrity relied on a simple premise: place students under the watchful eye of an invigilator, restrict their access to external resources and evaluate their output. But as assessments increasingly move online, that model has steadily broken down. Students now have access to a growing ecosystem of digital tools, many of which are difficult or impossible for institutions to detect during live exams.

The question institutions now face isn’t simply how to catch students who cheat, but whether exam design itself needs to evolve. 

Naturally, this increases the pressure on institutions to crack down on misconduct while maintaining student accessibility and trust. However, cheating is a complex issue that cannot be solved by adding more invigilators or deploying more invasive surveillance technologies.

There are also clear legal and ethical limits with aggressive proctoring. Under GDPR laws, institutions must have legitimate grounds for monitoring students and collecting personal data, while the EU AI Act requires automated exam monitoring systems to be transparent, retain clear audit trails and allow meaningful human oversight.

As a result, UK institutions are caught between two competing pressures: protecting academic integrity while avoiding disproportionate surveillance that could undermine student privacy, trust and the assessment experience. Rather than doubling down on proctoring, institutions need to recognise that the nature of cheating has changed and that safeguarding academic integrity now depends on designing exams that make cheating less valuable in the first place.

The game theory of cheating

In an environment where students have access to more resources than ever before, cheating has shifted from occasional infractions to an open secret. Ofqual says mobile phone and smart device malpractice has been the most common form of cheating in every summer exam series since 2018. Russell Group universities have also reported a 186% increase in generative AI misuse during the 2024-25 academic year, with cases rising from 700 the previous year to more than 2,000.

Compounding the issue is the fact that institutions cannot rely on static solutions to address their cheating concerns. The edtech space has seen a sharp increase in the number of plagiarism detectors that promise they can identify misconduct with a reasonable degree of accuracy. However, these tools can only flag patterns generated by the current generation of large language models. As AI is constantly evolving and changing, algorithmic detection becomes increasingly unreliable and is extremely limited in this sense.

If a student believes a significant number of their classmates are using second devices, AI or answer-sharing networks to improve their grades, staying strictly within the rules feels like an inherent disadvantage; this fundamentally alters student behaviour through a toxic form of game theory.

The risk rises further when institutions lack the necessary processes to identify and prove misconduct. When cheating slips under the radar, it silently encourages students to continue cheating, rewards dishonesty and places those who act with integrity at the greatest disadvantage. Undetected cheating can also create a false illusion of competence, damaging the credibility of institutions and the value of accreditations.

To build these processes and audit trails, institutions must accept that no single proctoring method is 100% foolproof. Those that recognise these limitations and implement multi-layered procedures, combining fair student monitoring, thoughtful exam design and clear audit trails, will be the ones capable of withstanding this new era of invisible cheating, rather than those who accuse with limited evidence.

Tech-assisted cheating is a design issue, not just a surveillance one

Although AI-assisted cheating has become a major issue, it is not unique. The rise of search engines and then smartphones raised the valid argument that students will always have easy access to online resources.

During my time at medical school, before AI became mainstream, academic leaders managed a different version of this cheating epidemic. They realised doctors are now entering hospitals with devices in their pockets containing the latest clinical and treatment guidelines.

Was it then useful to only assess students in a controlled environment without these resources?

Our dean realised it wasn’t. Instead, they wanted to create exams where students could access certain materials under controlled conditions, shifting the focus from recall to how efficiently students can synthesise information and critically evaluate data to arrive at a safe clinical decision.

Generative AI is not a departure from this trend; it’s accelerating it. While closed-book formats are still crucial for measuring a student’s foundational knowledge, reverting to them permanently simply because of cheating concerns in other exam formats means institutions miss out on vital opportunities to evaluate a student’s academic capabilities and risk underdeveloping other skill sets.

The importance of embedding integrity into the delivery process

The future of secure online exams requires institutions to shift away from aggressive, invasive monitoring tactics and focus on thoughtful assessment design. Rather than banning online tools altogether, institutions could place a greater emphasis on integrating them securely within controlled exam environments.

In practice, students could be asked to solve a hyper-contextual problem whilst being provided with online resources under proctored testing conditions. This could involve a stripped-down web browser that gives students access to certain resources but blocks communication apps and unapproved websites. When this is combined with strict time limits, students are more likely to focus on the exam rather than wasting time trying to cheat.

Institutions could also allow students to use certain AI tools but under strict testing environments, where tokens are limited and all prompts, chats and interactions are logged as part of the official exam record. This approach allows institutions to maintain control over usage and allows them to tailor parameters on a per-exam basis.

By focusing on exam design, institutions can move away from grading a student’s final answer to better understanding their thinking process. In these models, because access to online resources is built into the delivery system itself, the immediate advantage that AI offers becomes much less valuable. Students are not only required to provide a correct answer; their judgement and reasoning are also taken into account, which AI cannot compensate for.

Where assessments are high stakes and open-book formats might not be suitable, institutions can also look to well-established models of academic defence that are common in PhD settings. This method combines written work with an oral examination, where students are expected to justify their reasoning to an examiner.

These examples illustrate that exam security can be achieved without resorting to heavy proctoring. When institutions embed integrity into exam design and delivery processes, cheating becomes not just harder but entirely a waste of time and effort for students. It bypasses the endless debate over whether a student has cheated their way to a correct answer, and instead focuses on a simpler metric: do you actually understand the answer you’ve submitted, and can you explain your logic?

Proctoring is only one part of the solution

Turning student laptops into surveillance hubs during an exam is a limited approach to exam security.

The future of exam security won’t be determined by who develops the most sophisticated surveillance technology. It will depend on whether institutions recognise that the very nature of cheating has evolved. As misconduct becomes increasingly invisible, institutions must move beyond a proctoring-first mindset and design exams where cheating holds no value, and genuine understanding is the only path to academic success.

When exams test how students think, rather than just what they produce, cheating becomes far less valuable and academic integrity becomes far easier to protect.

By Dr. James Gupta, CEO and Founder of Synap