AI Smart Glasses Are Breaking the Assumptions Behind Supervised Exams

Traditional exam supervision rests on a simple premise: the person in the room is the person producing the answer. AI-equipped glasses, earbuds and connected wearables make that premise harder to establish.

A camera embedded in ordinary-looking frames can capture a question; an outside helper or AI service can process it; and a lens display or audio channel can return information with little obvious movement. The concern is therefore larger than one new cheating device. It is that a conventional invigilated exam may no longer be reliable evidence that an answer was produced without external assistance. 46

The 2026 cases show the risk is already real

Recent incidents have appeared across several examination systems:

• University of Auckland: the university disciplined a student for taking AI glasses into a test after organisers had explicitly prohibited smart devices. The university said its examination staff were being trained to recognise the technology and respond to it. 48
• National Taiwan University: NTU disqualified an applicant from an entrance examination for using AI smart glasses and assigned a null score. Reporting said invigilators became suspicious after the candidate repeatedly lowered their head; the glasses’ frame was found to be hot during inspection. 49
• South Korea: two test takers were caught wearing AI glasses in separate May TOEIC sessions, and their scores were invalidated. Other reported cases involved national technical-qualification examinations; later reporting said at least five candidates had been caught in May and June, prompting plans to revise rules. 364032
• The SAT: the College Board’s current prohibited-devices rules explicitly bar smart glasses, phones, smartwatches, fitness trackers, Bluetooth devices such as wireless earbuds, and other wearable technology during testing and breaks unless a student has an approved accommodation. The explicit smart-glasses prohibition began in spring 2026. 1821

These are detected cases, not a measure of all attempted misuse. They do, however, establish that wearable-enabled cheating is operational across university, language-testing, professional-qualification and admissions contexts.

Why wearables are difficult to detect

Smart glasses are challenging because they can look like normal prescription frames while incorporating cameras, microphones and other electronics. An educator quoted by CQUniversity described a pathway in which glasses capture questions and return answers within seconds through audio or an in-lens display. 46

Detection often depends on clues rather than a universal screening method: unusual head position, repeated touching of a frame, a visible reflection or sound, or an examination staff member deciding to inspect a device. NTU’s case is revealing precisely because suspicion and physical inspection were necessary before the device was identified. 9

The problem also extends beyond glasses. Wireless earbuds, watches, rings and concealed phones can distribute parts of the same workflow: one device captures or transmits information, another receives it. A rule or visual search focused only on phones can miss the wider connected system.

Claims that a particular wearable setup can reliably achieve a top or perfect score should be treated cautiously unless independently tested under stated exam conditions. The reporting here supports confirmed attempts and sanctions; it does not establish a general success rate across exams.

Why a ban is necessary—but not sufficient

Clear bans are important. They set expectations, give invigilators authority to intervene and protect students who follow the rules. The College Board’s policy is a straightforward example of that boundary-setting. 18

But an enforcement-only strategy has limits:

• Wearables can become less visually distinctive over time.
• Intensive searches and constant surveillance are costly and can generate false suspicions.
• Students who wear prescription glasses or use approved assistive technology require fair, workable accommodations.
• Even strong room controls cannot by themselves demonstrate that every submitted answer reflects independent understanding.

This is an assessment-design problem as much as a device-security problem. A timed written response is increasingly weak evidence when AI can mediate recall, explanation and calculation outside the examiner’s view.

What universities can assess instead

The practical response is not to abandon all controlled exams. Institutions may still need assessments of unaided foundational knowledge. But high-stakes decisions should draw on multiple forms of evidence rather than a single easily mediated written sitting.

Useful options include:

1. Authentic supervised performance. Practical tasks, simulations, laboratory work, clinical exercises and case-based work can make judgement and method visible.
2. Short oral defences. A viva, demonstration or follow-up interview can test whether a candidate can explain choices, respond to a new scenario and correct an error.
3. Process evidence. Drafts, decision records, source evaluation, iterations, feedback and reflection can reveal how work developed—not merely what the final answer says.
4. Transparent AI-assisted assessment. Where AI use resembles real professional practice, assess disclosure, prompting, verification, limitations and responsible judgement instead of pretending the tool does not exist.
5. Triangulated evidence. Combine lower-stakes observations, supervised tasks and targeted checks so one compromised sitting carries less weight.

The key distinction is between assessing what someone can do without assistance and what they can do with tools responsibly used. Both can matter, but they are different claims and should be assessed transparently.

The wider privacy and governance issue

The same capabilities that threaten exam integrity—discreet cameras, microphones, image capture and networked assistance—also raise concerns about recording people without meaningful notice. Schools, universities and employers need clear policies for device-free areas, consent, recordings, storage and deletion, investigations, disability accommodations and proportionate searches.

Governments also face product-policy questions: whether camera-equipped wearables should have more conspicuous recording indicators, stronger default privacy settings, seller duties, disclosure requirements or restrictions on devices promoted for covert use.

The available reporting supports the exam-security cases and policy responses described above. It does not provide sufficient evidence for jurisdiction-specific legal conclusions in the UK, England and Wales, Norway or Australia. Privacy, surveillance, criminal and consumer-product rules differ by place and setting, so those questions require separate review rather than a single global legal verdict.

The bottom line

AI wearables expose a weakness in traditional supervision: seeing a student at a desk is no longer the same as knowing they worked alone. Explicit device bans and trained invigilators remain essential safeguards, but the durable solution is to make assessments demonstrate understanding, judgement and process—not simply produce an answer that a connected device could help generate.