As the technological and economic shifts of the digital age dramatically shake up the demands on the global workforce, upskilling and reskilling have never been more critical. As a result, the need for reliable certification of new skills also grows.
Given the rapidly expanding importance of certification and licensure tests worldwide, a wave of services tailored to helping candidates cheat the testing procedures has naturally occurred. These duplicitous methods do not just pose a threat to the integrity of the skills market but can even pose risks to human safety; some licensure tests relate to important practical skills like driving or operating heavy machinery.
After firms began to catch on to conventional, or analog, cheating using real human proxies, they introduced measures to prevent this – for online exams, candidates began to be asked to keep their cameras on while they took the test. But now, deepfake technology (i.e., hyperrealistic audio and video that is often indistinguishable from real life) poses a novel threat to test security. Readily available online tools wield GenAI to help candidates get away with having a human proxy take a test for them.
By manipulating the video, these tools can deceive firms into thinking that a candidate is taking the exam when, in reality, someone else is behind the screen (i.e., proxy testing taking). Popular services allow users to swap their faces for someone else's from a webcam. The accessibility of these tools undermines the integrity of certification testing, even when cameras are used.
Other forms of GenAI, as well as deepfakes, pose a threat to test security. Large Language Models (LLMs) are at the heart of a global technological race, with tech giants like Apple, Microsoft, Google, and Amazon, as well as Chinese rivals like DeepSeek, making big bets on them.
Many of these models have made headlines for their ability to pass prestigious, high-stakes exams. As with deepfakes, bad actors have wielded LLMs to exploit weaknesses in traditional test security norms.
Some companies have begun to offer browser extensions that launch AI assistants, which are hard to detect, allowing them to access the answers to high-stakes tests. Less sophisticated uses of the technology still pose threats, including candidates going undetected using AI apps on their phones while sitting exams.
However, new test security procedures can offer ways to ensure exam integrity against these methods.
How to Mitigate Risks While Reaping the Benefits of Generative AI
Despite the numerous and rapidly evolving applications of GenAI to cheat on tests, there is a parallel race ongoing in the test security industry.
The same technology that threatens testing can also be used to protect the integrity of exams and provide increased assurances to firms that the candidates they hire are qualified for the job. Due to the constantly changing threats, solutions must be creative and adopt a multi-layered approach.
One innovative way of reducing the threats posed by GenAI is dual-camera proctoring. This technique entails using the candidate’s mobile device as a second camera, providing a second video feed to detect cheating.
With a more comprehensive view of the candidate's testing environment, proctors can better detect the use of multiple monitors or external devices that might be hidden outside the typical webcam view.
It can also make it easier to detect the use of deepfakes to disguise proxy test-taking, as the software relies on face-swapping; a view of the entire body can reveal discrepancies between the deepfake and the person sitting for the exam.
Subtle cues—like mismatches in lighting or facial geometry—become more apparent when compared across two separate video feeds. This makes it easier to detect deepfakes, which are generally flat, two-dimensional representations of faces.
The added benefit of dual-camera proctoring is that it effectively ties up a candidate's phone, meaning it cannot be used for cheating. Dual-camera proctoring is even further enhanced by the use of AI, which improves the detection of cheating on the live video feed.
AI effectively provides a ‘second set of eyes’ that can constantly focus on the live-streamed video. If the AI detects abnormal activity on a candidate’s feed, it issues an alert to a human proctor, who can then verify whether or not there has been a breach in testing regulations. This additional layer of oversight provides added security and allows thousands of candidates to be monitored with additional security protections.
Is Generative AI a Blessing or a Curse?
As the upskilling and reskilling revolution progress, it has never been more important to secure tests against novel cheating methods. From deepfakes disguising test-taking proxies to the use of LLMs to provide answers to test questions, the threats are real and accessible. But so are the solutions.
Fortunately, as GenAI continues to advance, test security services are meeting the challenge, staying at the cutting edge of an AI arms race against bad actors. By employing innovative ways to detect cheating using GenAI, from dual-camera proctoring to AI-enhanced monitoring, test security firms can effectively counter these threats.
These methods provide firms with the peace of mind that training programs are reliable and that certifications and licenses are veritable. By doing so, they can foster professional growth for their employees and enable them to excel in new positions.
Of course, the nature of AI means that the threats to test security are dynamic and ever-evolving. Therefore, as GenAI improves and poses new threats to test integrity, it is crucial that security firms continue to invest in harnessing it to develop and refine innovative, multi-layered security strategies.
As with any new technology, people will try to wield AI for both bad and good ends. But by leveraging the technology for good, we can ensure certifications remain reliable and meaningful and that trust in the workforce and its capabilities remains strong. The future of exam security is not just about keeping up – it is about staying ahead.
The post Is Generative AI a Blessing or a Curse? Tackling AI Threats in Exam Security appeared first on Unite.AI.
