Anyone with little money can buy access to the Moscow surveillance system connected to tens of thousands of cameras, and check the stored shots over the past five days.
Forum vendors and message groups that also handle illegal data also provide search technology with face recognition technology.
To ensure safety in the city, Moscow has more than 175,000 CCTV cameras, most of which are installed at the entrances, and more than 4,000 located in crowded places.
In 2017, the Moscow mayor's office announced that facial recognition technology integrated with Russian police databases had been activated, and that it was pulling data from 3,000 cameras.
The city's website says the video surveillance system can be accessed by "employees of federal government agencies, the mayor of Moscow and their authorized officials, law enforcement and executive authorities."
Investigative software company MBKh Media has found that access to this technology and live broadcasts are being sold in secret forums and chat rooms.
The seller, Andrei Kaganski, who conducted the investigation, says that the sellers are Moscow police and government officials who can log into the integrated data processing and storage center known as "YTKD", a system that saves data from Moscow's cameras.
And everyone who wants to check the live broadcast from the camera gets a unique link to the "City CCTV" system, which connects to all public cameras in Moscow. This link operates for five days, says Kaganski.
It is the same period mentioned in the city's CCTV division to store snapshots of crowded places, shops and yards, where data from educational organizations is kept for 30 days.
Government officials or police officers sell their entry data to the system to provide unlimited access to all cameras at an acceptable price of 30,000 rubles ($ 470), according to Kaganski.
To test facial recognition features, the investigator provided a photo of him to the seller. The research re-sent 238 photos of people (male and female) with a similar look from 140 cameras, along with a list of titles and times they were taken on the camera.
Among the metadata available for each image, there was a poster indicating whether the person was regularly seen by that camera or if it was detected for the first time.
As for the accuracy of the results, none of the images returned were similar to the investigator’s photo. However, facial features were similar to the input and evaluation system found to be similar to 67%.
Kagansky says in his photo report, bad results can be explained by a limited number of cameras related to the face recognition algorithm.
The investigator says the vendors he was in contact with told him that the police had free access to this technology, without any legal justification required to conduct a search.
Finding sellers does not seem to be a difficult task as forums are indexed by search engines, so their content is easy to discover. This type of service is required by investigators, scammers, or individuals who trade in surveillance services.