Moscow’s local government has formally announced the deployment of facial recognition technology on a “city-wide” network of CCTV cameras. The system has been undergoing tests for close to a year, but the city’s Department of Information Technology today revealed new details of the project, including its licensing agreement with Russian startup NTechLab for the facial recognition software itself.

Speaking to The Verge, the chief information officer of the Department of Information Technology, Artem Ermolaev, said the project is still “somewhere between” testing and finished, but has already led to six arrests. “We connected to a federal list of [wanted criminals] for the past two months, and arrested six persons, who hadn’t been caught for several years,” said Ermolaev.

The system taps into Moscow’s network of 160,000 CCTV cameras, which cover the entrances to 95 percent of the city’s apartment buildings. However, due to the cost of deploying the technology, only a few thousand cameras are active at a time. The cameras scanning for faces can be changed on the fly, says Ermolaev, to target “regions most concentrated with crime,” or to surveil areas where suspects are known to frequent. The city is currently working with NTechLabs for ways to minimize costs and expand its coverage.

What’s particularly unclear at this point is to what degree the system is automated. Ermolaev says that there are “several ways” to use the technology. Sometimes, he says, the system scans for faces that match mugshots stored in a police database, and other times his team will screenshot footage from CCTV cameras and feed that into the facial recognition software manually. There are no exact figures for how many faces are scanned on a daily basis, but Ermolaev suggests it is somewhere in the region of several thousand.

The extent to which this process can be automated is an important limitation on how widely the technology will be deployed. NTechLab, the supplier of the facial recognition tech behind the system, already makes its FindFace software free for anyone to use a limited number of times a day. FindFace matches faces to profile pictures from Russia’s Facebook equivalent, VK, which was the subject of controversy last year after it was reportedly used to identify porn stars.

Scaling this sort of facial recognition system up to the city-wide level is also constrained by the quality of footage from CCTV cameras. The pictures they capture are usually grainy and the subjects far away, so although the Department of Information Technology claims it has access to a potential network of 160,000 cameras, it’s not clear how many of these produce usable images.

Speaking to The Verge, Ermolaev would not give specific figures for the accuracy of NTechLab’s software when it comes to analyzing CCTV imagery. However, he told Russia Beyond last December: “In such conditions obtaining recognition of even 60-70 percent of the images is extremely difficult. A 30 percent result is already cosmic.”

These limitations aside, it’s clear that facial recognition is slowly being adopted around the world. Police in the UK have run a number of trials, including at Notting Hill Carnival in London and outside football games. In China, facial recognition technology is put to all sorts of uses, from spotting criminals to dispensing toilet paper. Deploying the technology reliably across an entire city for the purposes of spotting criminals is still some way off, but it’s much closer than we think.