Sector · Public safety
The glasses are the same. What changes is who is in the database
Sixteen cases. Seven fugitives caught at a station gate in 2018, a thousand officers on a parade route in 2026, two Harvard students naming strangers on the street, and the agency that banned its own people from wearing them.
This is the sector where the technology is easiest to describe and hardest to argue about. A camera on your face compares what it sees against a list. Whether that is police work or surveillance is decided entirely by the list, and not at all by the glasses.
The oldest hard result is from February 2018, at Zhengzhou East railway station, during the Spring Festival travel rush. Railway police wore 33.4 g facial-recognition glasses from Beijing LLVision at four entry gates, matching passengers against an on-device database. The operation caught 33 people: 7 wanted fugitives, including for human trafficking and hit and run, and 26 travelling on someone else's identity document. The vendor claimed a 0.1 second match against a 10,000-person database in testing, which is a vendor number. And the argument arrived with the deployment: Radio Free Asia reported dissidents in Guangdong who expected the glasses to be pointed at petitioners and protesters, while a Beijing academic argued they serve ordinary public security. Both of those things were said at the time, so both are here.
What is running now
In China the glasses have become routine kit. Officers in Tianjin use them on patrol to read text and objects and to help find missing people. Traffic police in Jinyun County report roughly a 300% improvement in the efficiency of roadside key-vehicle checks; Haikou reports the run from identification to alert completing in about 30 seconds. Chengdu pairs them with humanoid robots and robot dogs, and Shenzhen's 2026 work plan names agency-wide deployment as a target. Every one of those figures comes from the departments themselves, through a single report, with no independent measurement.
In January 2026 Delhi Police put AI glasses combining facial recognition and thermal imaging on more than a thousand personnel along the Republic Day parade route, matching offline against an encrypted local database of about 65,000 records at a 60% similarity threshold, with an unmatched face framed green and a match framed red. Neither source we found reports how many people were flagged or arrested. That absence is the most important fact about the case.
Where it did not take
The United States mostly tried and stopped. Byron, Georgia ran a one-day trial of Google Glass with recording software, no facial recognition, no plate scanning, and declined to buy: short battery, extra bulk, and the cost of fitting prescription lenses for officers. Centerville, Georgia did adopt them, from March 2015, simply as hands-free body cameras. The NYPD beta-tested and, in the sources reviewed, never confirmed a rollout. The Netherlands ran a proper study with Leiden, Twente and the police academy, and reported positive early feedback without calling it a deployment.
Then it turned around
Through 2025 and 2026, 404 Media documented ICE and CBP personnel wearing consumer Ray-Ban Meta glasses on immigration operations in at least six states, including a raid outside a Home Depot in Los Angeles. In August 2026 ICE banned them: a memo signed by Acting Director David Venturella classified the glasses as unauthorised personal body cameras that could capture or transmit sensitive information, about a month after the agency committed 30.9 million dollars to conventional body cameras. Meanwhile a leaked DHS budget justification for FY2027 asks for 7.5 million dollars to build agency ICE Glasses with facial recognition, iris scanning and gait analysis against a biometric database reported at roughly 75 million records. That one is a funding request, not a product.
And the version anyone can build
Two Harvard students, AnhPhu Nguyen and Caine Ardayfio, wired consumer Ray-Ban glasses to the PimEyes face search engine and public-record lookups with a language model stitching the answers together, and named strangers on the street in real time, with home addresses and relatives. One subject confirmed the details were correct. They did not publish the code, and said large technology companies had held comparable tools back too. The ACLU, EFF, EPIC, Access Now and 75 other organisations then wrote to Meta demanding it drop a reported face-identification feature for Ray-Ban and Oakley glasses and disclose any discussions with ICE and CBP, naming protests, medical clinics and the safety of domestic violence survivors. At the time of writing Meta had made no public commitment to those disclosures.