In recent years, we at The Constitution Project have warned that adding facial recognition scanning to police body cameras poses serious risks that could undermine basic privacy and due process rights. Unfortunately, the time to prepare for these risks is running out. This week, the Wall Street Journalreported that body camera vendors are preparing body cameras with real-time facial recognition capabilities, and law enforcement agencies could potentially deploy them as soon as this fall.
Real-time facial recognition is especially concerning because it means that body cameras will continuously scan the face of everyone passing police officers on the street, and immediately log and relay data. Before adding real-time facial recognition to body cameras, it’s critical that departments and lawmakers implement necessary measures to avert the unprecedented mass collection of the identity and location of individuals in public.
Set Standards for Police Action that do not Depend upon Facial Recognition
A major issue law enforcement must confront before deploying facial recognition is its inaccuracy. It is well documented that despite its immense power, facial recognition technology is often wrong, especially when identifying racial minorities. Specifically, these systems are prone to generating false positives, in which the technology identifies a match (e.g., says a person on the street matches the face of a wanted criminal) when in reality the faces are of two entirely different people.
It’s not hard to see how this situation could spin out of control with real-time facial recognition on police body cameras. What if an officer’s camera misidentifies an innocent person as a dangerous fugitive at large, leading to a violent incident? Even a commonly accepted police use for facial recognition—searching for a missing child—could turn horribly wrong if a false positive leads an officer to confront a parent as an abductor. Body camera use has grown exponentially because many saw it as a means to improve community-police relations and reduce use-of-force incidents, but adding facial recognition could inflame these problems. Even if misidentifications do not result in use of force, a mere arrest has serious consequences for individuals. They can be detained, fingerprinted, and subject to strip searches—all merely because a computer program was wrong.
It’s critical that before police add real-time facial recognition to body cameras, they set proper limits on the degree to which officers can rely on the identifications provided by an imperfect system. At a minimum, facial recognition should not be allowed to serve as the sole basis for an arrest or any use of force. Officers should seek means of corroborating an identification, and in the event of conflict, base their decision about what action to take on the totality of circumstances rather than completely trusting the determinations of a facial recognition program. This principle is consistent with current practices: Departments such as the NYPD already require human review to confirm results when facial recognition is applied to crime scene footage. These measures are necessary not just to prevent improper conflicts and arrests, but also to avoid a perverse incentive to build systems that generate more false positives, which would give police more pretexts to stop or arrest people, while limiting their liability for those actions because the identifications were based on the technology.
Limit Facial Recognition Scans and Identifications to Serious Crimes
Another serious risk that facial recognition poses is giving police “arrest-at-will authority,” and this potential is greatest when real-time facial recognition is incorporated into body cameras. Arrests may be a common police function, but they usually occur in response to specific assignments or situations, rather than in a random or opportunistic manner.
In some municipalities, a huge portion of the population has active bench warrants for minor violations, such as unpaid parking tickets (which people often don’t know can lead to an arrest warrant). For example a 2015 Department of Justice investigation revealed that 16,000 out of the 21,000 residents of Ferguson, Missouri, had outstanding warrants.
A patrol officer may be able to keep an eye out for the faces on a most-wanted listed, but they can’t memorize tens of thousands of people with outstanding warrants for petty offenses. Facial recognition changes that: This technology can take a face and scan it against millions of photos in a second. With this tool, every officer could be notified whenever they encounter an individual with any outstanding warrant, no matter how trivial the offense, and have free rein to arrest them.
This creates serious risk of abuse, as The Constitution Project’s comprehensive report on police body cameras—whose signatories include both civil liberties advocates and former law enforcement officers—warned. This “arrest-at-will authority” could also be wielded to disrupt First Amendment-protected activities. Police could use real-time facial recognition to scan crowds at protests or political rallies, and then arrest anyone flagged for any potential offense—no matter how trivial. Fear of such abuse isn’t paranoid—we’ve already seen it attempted. In 2016, police scanned for and identified any individuals with outstanding warrants among those protesting police brutality in Baltimore, using a social media scraping software tool called Geofeedia. The platforms Geofeedia scraped its data from (Facebook, Twitter, Instagram) quickly cut off Geofeedia’s access to end the program, but with police using body cameras equipped with facial recognition, there is no such middleman to block misconduct. This will allow law enforcement to directly disrupt and chill participation in First Amendment protected activities.
The solution to these issues is simple: Facial recognition incorporated into body cameras should only be used in relation to an enumerated set of serious crimes. This would set an effective balance, preventing potential abuses stemming from overbroad use while still allowing a system to flag serious threats for officers. Limiting use of powerful technological tools to serious offenses has precedent. The Wiretap Act, which sets the foundation for law enforcement surveillance of phone calls and electronic communications, is only allowed to apply to a list of serious crimes. The government cannot wiretap everyone suspected of parking violations, and it shouldn’t be able to deploy mass surveillance across American cities for such minor offenses either.
Provide Oversight to Prevent Unfettered Location Tracking
A final risk is that real-time facial recognition in body cameras creates a new avenue for location tracking that is devoid of accountability or oversight. Currently, law enforcement location tracking is mostly conducted by tracking cellphones; the United States Supreme Court is currently reviewing whether this should require a warrant. However, even if the Supreme Court does not impose a warrant standard, cellphone location tracking still requires some court approval. Facial recognition and body cameras, which currently do not require court approval to use, could cut out this independent oversight entirely, circumventing a basic due process protection.
Given the sheer scale of use of police body cameras in populated areas, facial recognition could allow law enforcement to rapidly scan and locate anyone they desire, and track their movements. This would circumvent privacy rights and independent oversight. It could also chill sensitive activities: If someone sees an officer near a protest, house of worship, political rally, or medical facility, they might (and should) worry that their presence at that location (along with every other attendee) is being logged.
In order to prevent abuse and preserve due process, it’s vital that a court approve any use of body cameras with facial recognition for location tracking, just as court approval is currently a key component of oversight of other forms of electronic location tracking. The best specific rules and standards for this activity may become more clear in light of the Supreme Court’s ruling on cellphone tracking later this spring, but at a minimum police departments should begin preparing to incorporate independent oversight into any type of location tracking for body cameras with facial recognition. The technology is too powerful—and location information is too sensitive—for law enforcement to be unchecked in its use.
There are a variety of avenues towards setting effective policies for body cameras. Some police departments have directly stepped up and adopted effective internal guidelines. In other locations, cities have established rules to ensure body cameras provide accountability rather than overbroad surveillance. State legislatures have also set limits to stop body cameras from becoming too pervasive as a surveillance tool. Individuals should consider engaging at all of these different levels of government, but now is the time to act. If we do not, we could soon be in a world where the government has an eye on every street corner, with little oversight or accountability about how it uses this immense power.