On AI glasses and wearable AI in assessment

Synthesis: Corbin, Sharpe, and Dawson (2026) argue that AI-enabled smart glasses and the wider ecology of wearable AI erode the two conditions on which the physical exclusion of AI from assessment depends: separability, the treatment of AI use as a discrete, bounded act, and observability, the assumption that such use leaves detectable traces. They name the resulting condition dual transparency — AI simultaneously incorporated into the wearer's activity and producing no reliable external signal for observers. The paper follows the consequences through assessment integrity practice, from the structural weakness of prohibition to a regime of bodily adjudication whose costs fall hardest on students with disabilities, chronic conditions, and religious dress practices. Because interactive orals and invigilated exams lose their evidential confidence, the paper frames wearable AI as a second wave of the AI challenge, qualitatively different from screen-based GenAI.

Key Findings

1. Physical exclusion rests on two conditions. Separability assumes AI use is discrete and bounded — a before, a during, and an after — while observability assumes such use leaves artifacts or behaviors that are, in principle, identifiable and therefore governable.
2. Wearable AI is dually transparent. It is phenomenologically transparent because a worn tool can be incorporated into the wearer's sense of bodily capability, and observationally transparent because its operation produces no reliable external signal.
3. Dual transparency is the product's organizing principle. Ray-Ban Meta's smart glasses passed over one million units sold by May 2024, Even Realities markets discreet control — "Control without anyone noticing" — and devices are priced from as little as $40 USD.
4. Uptake, not cheating, drives the governance problem. The incorporated functions that make wearable AI hard to govern — live captioning, real-time translation, contextual retrieval — make it genuinely useful, so students will wear it routinely for reasons unrelated to assessment.
5. Prohibition fails on its own terms. Bans from the College Board, the University of London, and the University of Iowa add a rule to an unchanged assessment design, working only where compliance can be inferred or non-compliance detected — otherwise they produce an enforcement illusion.
6. Enforcement drifts toward bodily adjudication. Once inspection begins, scrutiny shifts from the artifact to the body, asks whether a person is performing authentically, and presses hardest on those whose relationship to worn objects is least optional.
7. The second wave differs structurally, not by degree. Screen-based AI is episodic by architecture, requiring a turn away and a return, whereas wearable AI supplies the structural conditions for cognitive integration, so tasks built on separation stop making ongoing sense.

Separability, observability, and the trade-off logic of exclusion

AI and assessment is, the authors argue, a wicked problem whose competing goods cannot be fully reconciled, so institutional responses are trade-offs rather than solutions. Drawing on their own earlier work, they describe institutions retreating from the hard conceptual boundary between acceptable and unacceptable AI use to the one boundary that looks physically enforceable: separation of student from technology at the point of performance. Invigilated exams, supervised performances, and interactive orals all carry this logic, and the oral has been championed as a backstop that reliably demonstrates true understanding. The paper grants that controlled formats are not costless — timed exams can privilege recall and disadvantage students with language barriers, caring responsibilities, or test anxiety, while orals can reward fluency and confidence unevenly — but argues those costs were accepted because they purchased evidential confidence and fair, inclusive conditions. That purchase depends on two preconditions, separability and observability, and the paper's target is whether they still hold. Assessment that cannot attribute work to a person loses its certification function.

Why wearing the technology changes the tool

The paper's claim is not that wearable AI is a better cheating device but that bodily location changes the relation between user and tool. Drawing on phenomenology and the body schema, it uses Merleau-Ponty's cane: a novice perceives the cane itself, while an experienced user perceives the pavement through it, because the tool has been incorporated into a lived sense of capability. Worn AI has the structural conditions for that incorporation, because it operates within the wearer's perceptual field and responds contextually rather than through a separate interface. A student using glasses for real-time translation does not experience themselves as consulting a translator; they experience conversing, so the technology becomes a hybrid process rather than an external aid. Screen-based AI on a laptop or phone is architecturally different: it requires an attention shift out and back, so even routine use stays episodic. The first condition of exclusion, separability, therefore loosens.

Observational transparency and the second wave

The second dimension concerns what observers can detect. A cane is transparent to its user but publicly visible in operation; a car is transparent to its driver but regulated on public roads. Screen-based AI, even concealed, keeps a behavioral signature — a gaze drop, an attention shift, a posture oriented away from the task — which is exactly what device surrender and invigilation rely on. AI glasses eliminate that signature, which is also why they have succeeded at covertly recording private interactions: an examiner cannot see when the AI is active, what was asked, or what was returned. Extending the same logic to earbuds, pendants, rings, and AI-enabled hearing aids makes identification harder still. The authors call wearable AI a second wave, distinct in three ways: the property is structurally new, the same transparency that resists governance drives routine uptake, and where screen-based AI asked whether students delegate work, the second wave asks whether they are developing capacities that are hybrids of themselves and AI — a question that connects to work on metacognitive offloading.

From prohibition to bodily adjudication

Institutional responses so far follow the first wave's pattern: the College Board prohibits smart glasses during testing, the University of London bars them from time-limited assessment, and the University of Iowa advises instructors to state that AI wearables are prohibited. These are discursive responses — rules added to unchanged designs — and they fail on their own terms, because a rule that cannot be enforced produces an enforcement illusion rather than evidential confidence. The next step is enforcement, and the paper shows why it cannot stop at phones and laptops. Hearing aids are increasingly AI-enabled by default, and smart glasses already track health and deliver interventions, so the line between assistive and augmentative does not map onto the technology. Inspection therefore becomes adjudication of the body, requiring students to prove that their glasses are corrective and their medical devices authentically therapeutic. Eaton and colleagues note that the burden of judgment then shifts to instructors relying on intuition and behavioral cues, and the costs fall on equity-group students — those with disabilities, chronic conditions, and sensory impairments, and students whose religious practice involves head or face covering — who already navigate a "house of cards" of disclosure and justification. The paper notes that a student with a hearing impairment receiving live captions without disclosing a disability is a genuine accessibility gain. The object of suspicion shifts from the essay to the body, and detection of the artifact gives way to judging the person.

What this means for practice

• Instructors. Stop treating an in-room ban as assurance: if a rule cannot be verified, redesign the evidence the task produces rather than adding another declaration to unchanged conditions.
• Assessment designers. Build attribution into the task itself — follow-up questioning, on-the-spot variation, justification of method — because the paper shows physical separation is no longer a reliable source of validity evidence.
• Administrators. Audit any policy that now requires inspecting worn objects: it converts equipment checks into judgments about students' bodies and, on the paper's account, lands that burden on disabled, chronically ill, and religiously observant students.
• Policymakers. Reconsider whether controlled formats still purchase the evidential confidence they once did; if separability and observability no longer hold, policy should move from exclusion toward task and program redesign.
• Researchers. Treat incorporation as an open question: the paper explicitly flags whether hybrid capacities help or harm learning as unresolved and calls for empirical study of interactive orals in particular.

Limitations

• The paper collects no empirical data. It is a conceptual and theory-building argument, and its central claim is explicitly conditional: if wearable AI achieves the uptake the authors expect and exhibits dual transparency, then physical exclusion cannot hold.
• The authors state plainly that whether incorporation is good or bad for learning is an open empirical question, and that the second wave may produce new capacities or dependence that collapses once the tool is removed.
• The strongest conclusion is reserved for the invigilated exam; the paper says the interactive oral "is less certain, requiring further empirical exploration" and claims at most that it is no longer immune to AI.
• Support for detectability comes from institutional policy documents and manufacturer marketing rather than any measured study of covert AI use in assessment settings; the paper reports no study of students actually using glasses in an exam.
• The argument leans on an analogy with the first wave of generative AI and on a projected trajectory, since the authors note current AI glasses are at "the point of maximum institutional leverage" and expect movement toward greater bodily integration and invisibility.

Citation

Corbin, T., Sharpe, S., & Dawson, P. (2026). On AI glasses and wearable AI in assessment. Assessment & Evaluation in Higher Education. Advance online publication.