You can ban the glasses. You can't ban the future.
Schools cannot solve AI-enabled cheating with device bans or detection tools, because the technology gets smaller and cheaper faster than enforcement adapts. The durable response is redesigning assessments so AI cannot shortcut them, and that depends on training teachers in AI-era assessment design.
Now, the story behind that answer.
In May and June 2026, test-takers in South Korea were caught, twice, wearing AI-powered smart glasses during a high-stakes English proficiency exam. Similar cases surfaced in Taiwan and China. The glasses look ordinary. They rent for about US$6 a day. And when researchers at the Hong Kong University of Science and Technology connected a pair to a frontier AI model and pointed them at an undergraduate engineering exam, the AI scored in the top 5 percent.
Regulators responded the way regulators do. The College Board banned smart glasses from the SAT. The UK exam regulator Ofqual issued a public warning about GCSE and A-level integrity. Education ministries across East Asia are rewriting exam protocols mid-year.
Here is the question none of those press releases answers: what happens when the glasses become contact lenses, or earbuds indistinguishable from hearing aids?
Why doesn't banning devices or using AI detection work?
Every response so far has been a detection response: confiscate devices, install scanners, watch students more closely. It is the same playbook schools ran when calculators arrived, then phones, then ChatGPT, and the pattern is always the same. The technology gets smaller, cheaper, and harder to detect faster than enforcement can adapt. Detection is a static defence against a moving target, and schools that build integrity strategy on it are buying the losing side of an arms race, annually, at renewal pricing.
The numbers show how far behind enforcement already is. Ofqual data showed device-related malpractice accounted for over 44 percent of all student misconduct in the UK's summer 2025 exams, before smart glasses went mainstream.
And when cheating happens at scale, even elite institutions have no playbook. In one widely reported case this month, a Brown University economics professor compared his class's take-home midterm (average score: 96) with the in-person final (average: 48.6, a historic low for his course). Nineteen students failed. Some went from 100 on the midterm to zero on the final. The professor described the university's institutional response as meek. If an Ivy League university with near-unlimited resources wasn't ready, what does readiness look like for a school in Manila, Kuala Lumpur, or Jakarta?
What does an AI-proof assessment look like?
In East and Southeast Asia especially, a single exam can decide a student's university, career, and social standing. When stakes are that high and the tools are that cheap, cheating is not a moral anomaly. It is a predictable response to assessment designs that AI has made obsolete.
An assessment that smart glasses cannot defeat is not a harder version of the same test. It is a different kind of task:
- Oral defence: students explain and justify their answers live.
- Process-based assessment: drafts, revisions, and reasoning are graded, not just the final artefact.
- Contextual application tasks: problems requiring local context, personal experience, or collaboration that no model can fake.
- Paired supervised and open tasks: in-class creation combined with take-home reflection, so teachers see both what students produce under supervision and how they think when tools are available.
None of this is exotic. All of it is teachable. And every one of these approaches depends entirely on the classroom teacher's ability to design, run, and grade it.
Why is teacher training the real exam security strategy?
A detection tool is something a ministry can buy. Assessment redesign is something only a prepared teacher can do. That is the gap.
Good teacher training treats assessment not as a policing problem but as a design problem, one of the core areas where teachers need structured, progressive upskilling. A teacher who understands what AI can and cannot do can write a task AI cannot shortcut. A teacher who has never been trained on AI is left guarding the door against glasses.
The countries pulling ahead have figured this out. Singapore has committed to AI training for teachers at every level, including those still in training, by 2026. The Philippines has moved toward a graduated national AI policy framework rather than a blanket ban. The countries falling behind are still spending budgets on detection software while their teachers improvise alone.
What should school leaders do about AI cheating this term?
Audit your assessments, not your students. For every high-stakes task in your school, ask one question: could a student with an AI assistant complete this without learning anything? If yes, the task needs redesign, whether or not anyone is cheating yet.
Treat integrity as curriculum, not discipline. Students use AI because it is cheap, useful, and everywhere. Schools that channel that into transparent, taught, assessed AI use produce students who can work with these tools. Schools that only ban produce students who are good at hiding.
Invest in teachers before scanners. A scanner at the exam hall door buys one exam season of safety and costs money every year. A teacher trained in AI-era assessment design compounds across every class they ever teach.
The smart glasses will keep getting smarter. The only durable answer is teachers who are smarter about assessment than the glasses are about answers. That is not a technology problem. It is a professional development problem, and it is solvable.
Frequently asked questions
Can schools reliably detect AI smart glasses in exams?
Not for long. Current devices already pass as ordinary glasses, and the hardware is shrinking toward earbuds and lenses. Physical detection buys time, not security. Assessment redesign is the only defence that does not expire with the next device generation.
Is banning AI in schools effective?
Bans reduce visible use, not actual use. Surveys consistently show the large majority of students use AI regardless of policy; bans mainly push use underground where it cannot be guided or assessed. Graduated policies that teach disclosure and appropriate use outperform blanket bans.
What kinds of assessment can AI not fake?
Live oral defence, supervised in-class creation, process-based grading of drafts and reasoning, and tasks anchored in local or personal context. These assess thinking rather than output, which is the part AI cannot substitute.