Clearview AI carved out a market niche for itself as a provider of facial recognition tools for law enforcement agencies that find the technology challenging to implement on their own. The company’s plug-and-play surveillance capability entices government users with free trial periods and a database of billions of faces scraped without permission from social media. According to a new report, the technology has been used by more agencies than previously disclosed, sometimes without authorization. The report may not be complete, since many police departments belong to networks for sharing resources.
“BuzzFeed News has developed a searchable table of 1,803 publicly funded agencies whose employees are listed in the data as having used or tested the controversial policing tool before February 2020. These include local and state police, US Immigration and Customs Enforcement, the Air Force, state healthcare organizations, offices of state attorneys general, and even public schools,” the publication noted this week. The data was leaked to Buzzfeed by a source whose identity is being kept secret.
Uses to which the tool was put included searches for protesters, criminals—and friends and family members. Inappropriate searches on acquaintances could have been predicted by anybody aware of the abuse of official databases for curiosity and personal gain. “Police officers across the country misuse confidential law enforcement databases to get information on romantic partners, business associates, neighbors, journalists and others for reasons that have nothing to do with daily police work,” the AP reported in 2016. A massive facial recognition database is an enormous temptation for unscrupulous government employees already accustomed to misusing such tools.
Aware of such problems, Clearview AI rolled out features last fall designed to make searches more easily auditable to cut down on misuse. But implementation of the controls requires internal monitoring by law enforcement heads. That’s dependent on policy and on leaders actually knowing that officers are using the technology.
“In many cases, leaders at these agencies were unaware that employees were using the tool; five said they would pause or ban its use in response to questions about it,” Buzzfeed’s reporters added. But some organizations—such as the New York City Police Department—appear to be hiding deeper relationships with Clearview AI behind claims of ignorance.
Unofficial and even unauthorized use of a facial recognition tool is possible partially because the company offers free 30-day trials to anybody “employed by a federal, state or local law enforcement organization” who says that they “have received authorization from your supervisor at that law enforcement organization to request trial access to Clearview AI.” The requirement for supervisor approval was added last year.
The list may not even capture the full range of use of Clearview AI’s facial recognition technology since many smaller departments tap into larger agencies and resource-sharing networks.
“My agency does not have any type of facial recognition software,” Charles Wynn, police chief for Chino Valley, Arizona, told Buzzfeed when asked about his small department’s place on the list of Clearview AI users. “If we have a need for it we send the cases to either the Arizona Department of Public Safety or Rocky Mountain Information Network. I have double checked with all my investigators, including the ones assigned to off-site task forces and no one is using any software programs outside of the two intelligence agencies I mentioned before.”
The Arizona Department of Public Safety admitted to trying Clearview AI’s facial recognition tool, but claims it is no longer in use. Rocky Mountain Information Network, also listed as a user and one of six federally funded regional centers in the Regional Information Sharing Systems (RISS) Program, didn’t respond to queries. Serving well over 9,000 law enforcement agencies across the country, RISS is intended to make tools such as facial recognition available to agencies that might find implementing them on their own to lie beyond their needs and resources. Departments tapping into services offered through RISS wouldn’t necessarily appear on Buzzfeed’s list.
Clearview AI isn’t the only vendor of facial recognition software out there (Motorola-owned Vigilant is also a major player, though it’s better-known for license-plate recognition) but the more than 3 billion faces in its database give it an important edge over competing services. Even the FBI boasts “only” 640 million or so faces (as of 2019) against which to match images.
But the FBI built its database from public records, such as driver’s license repositories, and was called out for doing so. “[T]he FBI’s face recognition apparatus continues to balloon, threatening our fundamental liberties,” the ACLU warned after House Oversight Committee Hearings two years ago. Clearview AI, on the other hand, populated its vast database by scraping images from social media services without the permission of either the posters or the hosting companies. The company’s activities drew protests from Facebook, LinkedIn, Twitter, and YouTube, as well as a lawsuit over privacy violations in California following on earlier legal action in Illinois.
“Technology company Clearview AI’s scraping of billions of images of people from across the Internet represented mass surveillance and was a clear violation of the privacy rights of Canadians,” according to Daniel Therrien, Privacy Commissioner of Canada. The company subsequently withdrew from the Canadian market.
Facial recognition, like any automated means of identifying and tracking people, is something of a holy grail for cops and intelligence community types. The technology’s accuracy has improved, too, especially during the pandemic as algorithms have been refined to focus on eyes and noses unconcealed by facial coverings.
“Without masks, median system performance demonstrated a ~93% identification rate, with the best-performing system correctly identifying individuals ~100% of the time,” the Department of Homeland Security boasted in January. “With masks, median system performance demonstrated a ~77% identification rate, with the best-performing system correctly identifying individuals ~96% of the time.”
Clearview AI isn’t alone as a provider of surveillance technology to law enforcement. Agencies have purchased cell phone location data from marketing firms and telecommunications companies to track people’s movements. The surveillance state isn’t yet ubiquitous, but it’s increasingly available as a plug-and-play solution for any cop interested in a free trial period.