CAICT Releases China's First Voluntary AI Glasses Privacy Code

China's AI glasses sector has grown fast enough to outpace its own norms. On June 26, the South China Morning Post reported that the China Academy of Information and Communications Technology (CAICT), a research institute under the Ministry of Industry and Information Technology, released what is described as the first coordinated industry effort in China to address privacy risks from AI-enabled smart glasses.

The immediate catalyst was Rokid, a Hangzhou-based tech company whose user community became the focal point for covert-filming complaints. Users posted videos on Rokid's own forum showing strangers recorded without their knowledge on subway trains, in parks, on beaches, and in shopping centers. One viral video depicted a flight attendant unaware she was being filmed. Rokid removed several videos and suspended accounts earlier in June but had not responded to requests for comment by the time of the report.

The new code asks manufacturers to follow a minimum data collection approach, display visible indicators when recording is active, and secure explicit consent before capturing anything. As Gizmodo noted, LED recording indicators are already common in both Chinese and US smart glasses but lack legal requirement in either country. The CAICT code does not change that: it is entirely voluntary and carries no enforcement mechanism.

That gap is the honest caveat. A voluntary standard in a market moving this quickly is a statement of intent, not a constraint. What the reporting does not address is whether Beijing plans to give the code statutory teeth, or how consent would be operationalized in crowded public spaces where incidental capture of bystanders is nearly unavoidable. The code also does not appear to tackle real-time AI facial recognition, which represents a larger privacy surface than recording alone.

For practitioners building on wearable AI platforms, the direction is worth watching. If China moves to make these requirements mandatory, the consent and data-minimization provisions become design constraints, not just best practices. Hardware makers that build compliant recording indicators and minimal data pipelines into their products now will have less to redo later.