AI Smart Glasses Are Forcing Exams Beyond the Proctoring Arms Race

AI-equipped glasses are turning a familiar object into a potential exam interface: a device can capture material in front of the wearer, connect to external assistance or an AI system, and return information through a subtle display or audio channel. The concern is not that every student will use them this way. It is that conventional invigilation depends on keeping outside help separate from the candidate—and on being able to observe prohibited conduct. Wearable AI challenges both assumptions. 117

Why smart glasses change the cheating problem

A phone normally requires a student to retrieve, view or operate a visible object. Smart glasses and related wearables can make that interaction far less conspicuous. Research on wearable AI identifies risks across glasses, watches and sensor-based devices, including real-time retrieval, summarisation, translation and coaching through discreet interfaces. 17

The consequence is a shift from a simple prohibited-item problem to an evidence problem. Academic-misconduct processes generally need observable behaviour and defensible evidence. Yet glasses may resemble ordinary eyewear, while wearable technology can also serve legitimate medical or accessibility purposes. An academic analysis warns that attempts to physically exclude AI under these conditions risk increasing bodily scrutiny and placing disproportionate burdens on students with disabilities, health conditions or religious dress practices. 1

The capability is no longer hypothetical

A reported Hong Kong University of Science and Technology demonstration used Rokid-based smart glasses connected to GPT-5.2 on an undergraduate Computer Communication Networks exam. The system reportedly scored 92.5 out of 100, placing in the top five of a class of more than 100 students. 45

That result should be read carefully. It is a demonstration of what a particular device-and-model setup could do on one assessment, not proof that AI glasses will score similarly across subjects, exam formats or future models. But it illustrates the core vulnerability: when an assessment rewards answers that a networked model can retrieve or generate quickly, a wearable interface can make outside assistance difficult to distinguish from unaided work.

Cases and policy responses are spreading

Institutions and exam administrators are already responding to reported incidents:

• The University of Auckland disciplined a student who wore smart glasses in a test after organisers had explicitly prohibited smart devices. 2
• South Korea recorded five cases involving AI-powered smart glasses in national technical-qualification exams over just more than two months, prompting preparation for changes to examination rules. 18
• China required screening of all glasses for its national college-entrance examination, while South Korean education authorities discussed stronger prevention measures for major exams. 6

These measures can reduce immediate risk, especially where rules clearly identify smart glasses and other connected wearables as prohibited. But they do not solve the underlying issue: the technology is becoming less noticeable, while aggressive inspection can create fairness, privacy and accessibility problems. 119

Why experts see a “checkmate” risk for conventional exams

The most serious warning is not that cheating becomes impossible to catch in every case. It is that the confidence afforded by a traditional closed-book room may no longer match what the room actually proves.

Wearable AI weakens two foundations of conventional invigilation:

1. Separability: the idea that the candidate can be reliably separated from external knowledge sources and assistance.
2. Observability: the idea that an invigilator can see meaningful signs of misconduct.

An assessment paper on wearable AI describes this combined challenge as “dual transparency”: the user may appear transparent to the invigilator while the assessment itself becomes transparent to the device. It argues that physical exclusion alone is unlikely to restore assessment security. 1

The risk also extends beyond glasses. Smartwatches, earpieces and other wearable sensors can offer discrete assistance, and academic researchers note that regulation has lagged the adoption of such devices. 17 Future, still more inconspicuous display technologies would make a strategy based solely on visual detection even harder to sustain.

The wider stakes: privacy, credentials and equal treatment

Wearable AI raises concerns beyond academic dishonesty.

Privacy and consent. Devices with cameras and microphones may capture classmates, staff, exam material or bystanders without meaningful notice. The same features that enable covert assistance can also enable covert recording. 1

Trust in qualifications. In technical and professional assessments, undisclosed live assistance can undermine the claim that a successful candidate personally demonstrated the required competence. South Korea’s reported cases have moved the issue into regulatory reform for qualification examinations, not merely campus discipline. 18

Accessibility and procedural fairness. A blanket suspicion of anyone wearing glasses can be discriminatory and difficult to justify. Policies need clear accommodations and procedures that target device capability and unauthorized access without treating ordinary eyewear or assistive technology as evidence of wrongdoing. 119

What schools and testing bodies can do now

Near-term controls still matter. Exam rules should explicitly cover smart glasses, watches, earpieces and other connected wearables; explain how devices must be stored; state consequences in advance; and give invigilators consistent reporting procedures. Screening and inspection should be lawful, proportionate and paired with a workable accommodation process. 618

But detection should be treated as a safeguard, not the entire assessment strategy. A ban that cannot be applied fairly or reliably may create an illusion of security rather than trustworthy evidence of learning. 1

The longer-term answer: assess understanding, not just output

The strongest response is to redesign high-stakes assessment around evidence that is harder to outsource invisibly. Depending on the learning goal, that can include:

• oral defences or short follow-up questioning;
• supervised practical demonstrations;
• staged coursework that shows drafts, decisions and revisions;
• authentic cases with local or personalised constraints;
• process evidence, such as working notes and reflections; and
• tasks that require students to explain, adapt or defend their reasoning.

These formats are not cheat-proof, and they demand more time and training than a standard written paper. Their advantage is different: they make the learner’s own judgment, process and performance part of what is being assessed. Researchers increasingly argue for this shift away from technology prohibition alone and toward assessment designs that remain meaningful in an AI-present world. 11719

The practical conclusion is not that exams are obsolete. It is that institutions should stop assuming an invigilated room automatically proves independent work. Clear device rules, privacy-respecting enforcement and assessment designs that directly test competence will be more resilient than an endless race to spot the next hidden interface.