Untargeted Facial Recognition is Unvetted and Unsafe: New Orleans City Council Should Reject It
Part one of this two part series regarding New Orleans’ dragnet facial recognition program is available here.
The New Orleans City Council is set to vote on an ill-conceived ordinance that would facilitate the use of untargeted facial recognition surveillance by the New Orleans Police Department (NOPD). Recent reports from the Washington Post have revealed that not only has the NOPD intentionally misled the public for years about its use of facial recognition, but it has also relied on an unauthorized partnership with a private organization to violate existing restrictions on the use of facial recognition.
The use of untargeted facial recognition by law enforcement is flawed by design, and may have already caused improper police stops, investigations, and even arrests of innocent people. But because the program has been cloaked in secrecy — with the NOPD refusing to keep records on alerts the system triggered and how officers responded — whatever harms it has caused remain hidden. By touting a small set of success stories while sweeping a potential litany of failures under the rug, proponents of this system are seeking to exploit a distorted view of untested, unvetted, and unsafe AI mass surveillance. The New Orleans City Council should not be fooled by this effort — it should reject the proposed ordinance and impose stronger restrictions on the NOPD’s use of facial recognition.
Current New Orleans Law Restricts the NOPD’s Use of Facial Recognition, But the NOPD Has Attempted to Exploit Its Loopholes
New Orleans was once at the forefront of nationwide efforts to ban police use of facial recognition. But in August 2022, the City Council replaced its ban to allow the NOPD to use facial recognition to investigate 46 serious, mostly violent, crimes. The law does not require judicial approval, evidence of wrongdoing by individuals being scanned, or notice to defendants when the technology was used in investigations, making it far more permissive than guardrails many states have enacted.
Despite the generally lenient nature of the law, NOPD has deployed facial recognition far more expansively than lawmakers intended to allow. For years, the NOPD maintained an informal, unauthorized partnership with a private nonprofit called Project NOLA, which was founded by a former NOPD police officer and operates more than 200 facial recognition cameras (and thousands of standard security cameras) to constantly surveil the public in New Orleans. According to Washington Post reports, the facial recognition cameras are configured to continuously scan all faces on video and check them against a list of wanted suspects. If the software determines it has found a match, it sends real-time alerts to officers who have installed an app on their mobile phones. The officers can then follow-up on the alerts, such as by going to the locations to confront and arrest the subjects.
Using this system flies directly in the face of the current law, which prohibits facial recognition from being “used as a surveillance tool.”[1] But because the police did not set up the program, and did not formally request the use of facial recognition, they treated this partnership as a loophole permitting them to use results from Project NOLA’s facial recognition systems. This ignores the fact that current law prohibits the NOPD from using information derived from the use of facial recognition that was not authorized.[2]
Disturbingly, the NOPD omitted information about the alerts it received from Project NOLA from its mandated quarterly reports to the City Council, did not catalog alerts it received from Project NOLA, and did not record or retain information on the results of officers responding to alerts. By prioritizing the system’s secrecy over accountability, the NOPD thus made it impossible for the City Council to evaluate its efficacy. Project NOLA claims that it assisted with 34 NOPD arrests, but publishes no data that would permit verification of that claim. Meanwhile, any data about harms that facial surveillance systems are causing people in New Orleans — including whether its use has prompted the police to make improper stops, arrests, or uses of force — has been suppressed. This paints a picture of untargeted facial recognition’s impact that is not only incomplete, but deeply misleading.
The New Orleans City Council is Considering a Misguided Proposal to Legitimize Dragnet Surveillance
In response to these alarming revelations, the NOPD announced it has temporarily paused its partnership with Project NOLA and ordered a review of how its officers use Project NOLA’s systems. But now it has proposed an ordinance that would legitimize its shadowy workaround with Project NOLA and authorize the NOPD itself to use facial recognition in more circumstances.[3]
The proposed ordinance would permit the NOPD to accept tips generated from facial recognition matches from privately-operated facial surveillance systems where the operator has entered into an agreement with the NOPD. Such an agreement would require the operator to disclose information about the software it uses, the source of the database it uses to generate matches, and statistical data about the accuracy and efficiency of the system.[4] But no independent testing would be required. No standard of accuracy would have to be met. No measures of racial, gender, or other bias would have to be disclosed. Project NOLA could meet these lax requirements even if 9 out of 10 alerts it issued were false alarms (as occurred in pilots of untargeted facial recognition systems tested in the United Kingdom).
Worse still, the proposed ordinance would relieve NOPD from reporting key information about the use of facial recognition in police investigations when such use stems from unsolicited tips that Project NOLA or another facial recognition operator provides. This includes information identifying the police officer who used the tip to identify a person to stop, arrest, or detain as part of their investigation; that person’s age, gender and race; and instances in which such a tip was provided, but an arrest was not made because, for example, the tip identified the wrong person. Under the proposed ordinance, none of this information would have to be reported when the tip was initiated by the third party operator rather than the result of a police request.
In this way, the proposed ordinance follows a pattern of the NOPD evading restrictions by outsourcing use of facial recognition to third parties. This “see no evil” approach deprives the City Council and the public of the ability to evaluate the system’s true efficacy, and prevents any New Orleans residents who are improperly stopped, arrested, or otherwise harmed by an untested AI surveillance system from seeking redress.
At the same time as it would allow police to avoid accountability, the proposed ordinance would quietly sneak in other expansions of the NOPD’s surveillance powers. For example, the proposed ordinance would add drug crimes and pickpocketing to the 46-crime list of permissible use cases.[5] These are massive expansions for non-violent crimes.
Furthermore, the proposed ordinance would authorize the use of facial recognition in broad circumstances that need not even relate to investigation of any of these crimes. For example, facial recognition could be used to “locate an individual for which a valid arrest warrant exists.”[6] In 2019, there were arrest warrants outstanding for one in seven adults in New Orleans, frequently stemming from minor nonviolent offenses that do not carry jailtime, such as panhandling. Even if the number of outstanding arrest warrants have decreased since then, an authorization to use facial recognition to find anyone under an arrest warrant (even for petty offenses such as unpaid traffic tickets) would risk facilitating the use of an arrest-at-will power over a huge portion of the population that would be highly susceptible to abuse.
The New Orleans City Council Should Reject Dragnet Facial Recognition
The untargeted facial recognition system the NOPD has used is untested, unreliable, and has likely caused false alerts and improper police action. And the dangers posed by misidentifications are only part of the problem. The use of untargeted facial recognition fosters a surveillance panopticon in which law enforcement could catalog every movement of every person. No one would be able to enter a place of worship, or engage in protected activities like public protest, without the threat of surveillance. Unfettered surveillance chills even our most basic speech and activities — people are less likely to protest, gather, or simply go about their lives without fear if they know they are being scanned, tracked, and flagged.
The NOPD has demonstrated time and time again that it cannot be trusted to not abuse dragnet surveillance tools, and yet the City Council appears to be considering an ordinance that would put facial recognition right back in their hands. Instead, the City Council should act decisively and reaffirm its commitment to civil rights by reimposing a categorical ban on untargeted, dragnet facial recognition.
What’s happening in New Orleans should also be a warning to other cities: outsourcing surveillance compounds the difficulty of controlling its abuse. Lawmakers, advocates, and communities across the country should reject this model.
[1] Sec. 147-2(g).
[2] Sec. 147-2(a)(1).
[3] The proposed ordinance would also make marginal improvements to current law. For example, the proposed ordinance would limit retention of data obtained through facial recognition to no more than 30 days unless part of an active and ongoing investigation, or as otherwise required by law. Proposed Ordinance, Sec. 147-3(g). It would also extend to other surveillance technologies the provision of current law establishing that facial recognition alone is insufficient to establish probable cause. This would be an improvement, but remains weak because surveillance technologies will never be the sole basis for probable cause because probable cause is determined from the entire context or the “totality of the circumstances.”
[4] Proposed Ordinance, Sec. 147-2(g).
[5] Proposed Ordinance, Sec. 147-2(d)(49). Current law permits use of facial recognition for investigation of crimes including armed robbery, first degree robbery, second degree robbery, simple robbery, and pursesnatching. Sec. 147-2(d).
[6] Proposed Ordinance, Sec. 147-2(f)(1).
