Why Exam Proctors Are Targeting Smart Glasses for Cheating Prevention

What a Pair of Glasses Can Do

In December 2025, a research team from Hong Kong University of Science and Technology (Prof. Zhang Jun and Prof. Meng Zili) connected a Rokid smart glass to GPT‑5.2 and let it take a university‑level computer‑network exam lasting three hours. The AI system submitted answers in 30 minutes and scored 92.5, surpassing 95% of the class (average 72, top 97.5 by a human).

Smart‑Glass Market Is No Longer Niche

In Q1 2026 the online smart‑glass market featured 275 brands. Rokid held the largest share at 21.2% with an average price of ¥3,178; Quark AI ranked second with 15.4% and ¥3,766; XREAL was third with 11.6%.

Current mainstream glasses support major AI models such as Tongyi Qianwen, Doubao, and DeepSeek, and offer camera capture, voice assistant, projection display, real‑time translation, and a “question‑solving” module that shows answers only to the wearer.

Even Offline, They Can Serve as Cheat Sheets

Beyond network‑dependent features, many glasses provide a “text” function that lets users preload TXT files onto local storage. Some devices store tens of thousands of characters, effectively turning the frame into an offline cheat sheet that works despite signal‑blocking measures.

How Cheating Methods Have Evolved

Early cheating relied on physical information transfer (hand‑written notes, hidden papers, answers written on restroom walls – e.g., a 2021 Jiangxi exam where answers were posted on a toilet door).

Subsequent generations introduced Bluetooth pens, Bluetooth watches, and micro‑earpieces, followed by phone‑based photo‑to‑answer transmission; a 2020 case resulted in a three‑year prison sentence and a ¥6,000 fine.

These incidents prompted stricter security (metal detectors, wireless‑signal shielding, RF detection), and now AI‑enabled glasses raise the stakes further.

Why Glasses Are Particularly Hard to Defend

Glasses cannot be confiscated from students with vision impairments, and many smart glasses visually resemble ordinary prescription lenses, some even accommodating real corrective lenses, making visual detection impossible. Traditional metal detectors cannot spot embedded cameras or transmission modules.

Current provincial measures require candidates to remove glasses under camera supervision for inspection, and exam staff receive specialized training to spot abnormal shapes or sizes. In 2026, “smart inspection gates” were added to specifically detect smart glasses, achieving 100% real‑time coverage.

AI Platform Countermeasures: Throttling

During the 2025 exam period major AI providers (Alibaba, ByteDance, DeepSeek, Tencent, Moonshot) disabled or limited question‑answering features. For example, Tongyi Qianwen blocked access from 08:00‑11:30 and 13:00‑18:30, matching exam slots, while DeepSeek imposed broader restrictions on any high‑school exam queries.

These blocks only affect cloud‑based services; on‑device inference and local storage remain untouched.

A Comparative Figure

In 2024 Fudan University’s NLP lab evaluated 13 large models on a high‑school math test; the best model (Doubao) achieved a 74.66% correct‑answer rate on objective questions, with many models failing on subjective items.

By the end of 2025, the HKUST experiment showed AI‑glass‑assisted students outperforming 95% of peers in a professional‑course exam, highlighting a rapid leap in AI answering capability.

Implications and Outlook

The core issue is not the technology itself but its deployment in high‑stakes, competitive assessments where speed and accuracy translate directly into unfair advantage. The current assessment paradigm—measuring recall under time pressure—becomes obsolete when AI can retrieve information instantly.

Upgrading proctoring methods merely patches a leaking dam; a deeper question remains: what should examinations aim to evaluate once AI tools become ubiquitous?

Reference sources: Jiemu News, IT Home, Tencent News, Sina Finance, Zhihu (Quantum Bit), People's Daily English, Global Times, Rest of World, Futurism, British Brief, Sixth Tone, Fudan University LLMEVAL report, HKUST Zhang/​Meng experiment, etc.