Facial recognition software has long been employed by governments across the world to assist in criminal probes and to screen people entering sensitive sites, from sports stadiums to customs checkpoints. An image captured from a surveillance video is compared to a photo database of known individuals, which increasingly includes billions of pictures scraped from social media. The software throws up a flag when, with some degree of probability, it spots a match.

But as other American cities have retreated from the technology or banned it altogether, Adams’ bullishness and the resources at his fingertips — the city is set to spend around $11 billion on the NYPD this budget cycle — stand to put New York at the forefront of an evolving national debate over safety, privacy and the racial and gender biases tied to the controversial software.

The NYPD has been using facial recognition technology for more than a decade, prompting at least six lawsuits and inspiring a New York statute mandating reams of public reporting. The same software used by the city, DataWorks Plus, led to the wrongful arrest of two men in Detroit.

Police officials have credited the tool with helping to solve murders, rapes and missing person cases, and they stress it is used fairly and only in a narrow capacity.

Now the mayor wants to go further.

“We will also move forward on using the latest in technology to identify problems, follow up on leads and collect evidence — from facial recognition technology to new tools that can spot those carrying weapons, we will use every available method to keep our people safe,” Adams, a retired police captain, said at a Jan. 24 press briefing.

A week later, he alluded to replacing metal detectors at public schools with new technology to scan students for weapons. A City Hall aide separately said the administration is exploring the use of infrared or thermal imaging cameras in the buildings.

The mayor’s zeal for high-tech policing tools was on display in his 15-page “Blueprint to End Gun Violence,” released last month on the heels of the shooting death of two NYPD officers responding to a 911 call of a domestic dispute. The document notes that new technologies used to identify suspects “will not be the sole means to make arrests, but as another tool as part of larger case-building effort.”

Those who have long warned about the department’s use of facial recognition technology were alarmed by the mayor’s plans.

“It just felt like the floor was dropping out from beneath me. To hear him this enamored with this sort of technology was incredibly disappointing,” Albert Fox Cahn, head of the nonprofit organization Surveillance Technology Oversight Project, said in an interview. “I was hopeful that he was someone uniquely positioned to push back against this sort of pseudo-science masquerading as public safety.”

The NYPD says it uses facial recognition to match surveillance or other crime-scene photos solely to its database of mugshots. But where law enforcement officers see an opportunity to close cases, watchdogs say the technology is largely unregulated, invades personal privacy and disproportionately targets people of color. The debate, which has played out in fits and starts in recent years, culminated in local legislation two years ago to mandate the NYPD regularly disclose details about its use of surveillance technology.

Adams is putting his faith in the software as he grapples with a scourge of gun violence — a matter he regards as the most profound problem facing the city, and one he discussed with President Joe Biden when the two met in New York last week.

“The city is at a critical turning point where we need to have more effective tools — plain and simple,” mayoral spokesperson Fabien Levy said. “The mayor is saying: Let’s study emerging technology and, if it’s effective, use it in a legal manner to make the streets safer while reasonably protecting the privacy of New Yorkers.”

Levy said the city would not use any software that results in biases against certain races or genders, though that is precisely what several facial recognition programs have been found to do. A 2019 study, for example, found that facial recognition programs across the country tended to be less accurate when analyzing Black or Asian individuals.

The NYPD’s use of facial recognition technology in particular was the subject of an unflattering report titled “Garbage In, Garbage Out,” penned in 2019 by privacy lawyer Clare Garvie.

“There are no rules when it comes to what images police can submit to face recognition algorithms to generate investigative leads,” Garvie wrote. “As a consequence, agencies across the country can — and do — submit all manner of ‘probe photos,’ photos of unknown individuals submitted for search against a police or driver license database.”

Garvie argued the pictures are sometimes poor quality or doctored through popular social media filters, rendering them unreliable evidence.

There have been at least three known cases of facial recognition resulting in a false arrest. In 2019, Detroit police used DataWorks Plus to identify 25-year-old Michael Oliver as a suspect in a theft case. An eyewitness later picked him in a lineup; he was arrested and falsely charged with a felony. Robert Julian-Borchak Williams was handcuffed on his front lawn in a Detroit suburb the following year, after the algorithm incorrectly identified him as a suspect in a theft case. That same year, New Jersey resident Nijeer Parks was jailed for 10 days after being falsely accused of stealing candy and attempting to hit a police officer with a car. He is now suing the township of Woodbridge.

NYPD has reported five misidentifications through its facial recognition technology between 2011 and 2017. But Supervising Attorney Jerome Greco of Legal Aid — which has represented clients in facial recognition cases and helped force the NYPD to disclose 58 private contracts with surveillance companies — said the criteria for what counts as a mismatch remains unclear.

And prior to a recent statewide change in the discovery process for criminal cases, the NYPD did not routinely disclose whether facial recognition was used at all. Thus, the true extent to which suspects have been falsely arrested or taken plea deals they otherwise might not have, based on faulty matches, will never be known.

“We’ve seen this already with stop and frisk,” he said, referring to the NYPD’s previous practice of conducting searches of mostly nonwhite New Yorkers with little pretext, which was found unconstitutional by a federal judge. “So you didn’t end up arresting someone, but you harassed the hell out of them. Why am I supposed to believe this is any different?”

Concerns like these led San Francisco, Seattle, Oakland and several other cities in America to ban the software altogether in recent years. Elsewhere, however, police forces have moved far beyond what Adams plans for New York City. Even the IRS has walked back a plan to use the software to confirm taxpayers' identities following backlash from legislators and privacy groups.

London is rolling out a program that will scan the faces of pedestrians in real-time and run them through facial recognition software in an attempt to match them with a database of wanted individuals. The process is similar in scope to China’s surveillance of its citizenry. London leaders have defended it as a necessary tool to get dangerous people off the streets, especially in the wake of terror attacks that have rocked the city.

Police officials in New York say their more tailored use of facial recognition has been an invaluable tool for combating crime. Most major offenses with the exception of murder were up in January compared to the same time last year, according to NYPD statistics.

“It’s been very helpful in a lot of cases in identifying violent suspects in crimes — quite a few sexual-related assaults and shootings where they were able to identify or get a lead to enhance the investigation through the facial recognition,” Paul DiGiacomo, president of law enforcement union the Detectives’ Endowment Association, said in an interview. “These anti-police groups keep trying to take away the tools that detectives use to investigate crimes and bring closure to some of these families that lost loved ones.”

Facial recognition is also more accurate than eyewitness identification, and like DNA evidence, can assist in solving cold cases and exonerating innocent people when applied with proper human oversight, according to former NYPD Commissioner William Bratton. He penned an op-ed defending the practice in 2020.

Bratton is now with global consulting firm Teneo and sits on the board of Shotspotter, a technology used by the NYPD and other American police departments to triangulate the sound of gunshots. He slammed what he considers a trend of left-leaning legislatures reflexively opposing arming police with new investigatory capabilities. He cited as an example concerns raised over the NYPD’s testing of a canine robot named Digidog — equipped with a camera and microphone, and designed to trot into situations hazardous to humans — that ended up forcing the agency to cancel the tryout last year.

“So many policies advanced by the progressive left are actually harming the people they are trying to protect,” Bratton, who served under former mayors Rudy Giuliani and Bill de Blasio, said in an interview. “And AOC’s comments on that robot are the case in point.”

Last February, Rep. Alexandria Ocasio-Cortez argued that the $75,000 device epitomized excess police funding and a willingness to infringe on the civil rights of New Yorkers with low incomes.

Her comments followed the New York City Council’s passage of the Public Oversight of Surveillance Technology Act, which monitors the police department’s use of technology. It was hastened by clashes between NYPD officers and protesters following the murder of George Floyd.

The bill requires public reporting. So far it has yielded 36 disclosures around the department’s monitoring of both cryptocurrency and social media, as well as a system that can locate phones or computers using a WiFi signal. It also mandates the city Department of Investigation annually produce an oversight report, which a spokesperson said is due in April.

Among the sponsors of the measure were Adrienne Adams, who now leads the 51-member Council, and Brad Lander, the recently-elected city comptroller with oversight of NYPD spending. Both are Democrats.

“When it comes to increasing the use of surveillance tools like facial recognition, we first have to ask – what safeguards are we using to protect civil rights?” Lander asked in a statement. “Facial recognition is notoriously inaccurate when it comes to identifying people of color generally and women of color in particular. We shouldn’t rush into expanding the use of technology that may increase the risk of racial profiling and false arrests.”

Lander’s predecessor, Democrat Scott Stringer, notified former Police Commissioner Dermot Shea in 2020 that he would begin compiling the NYPD’s contracts with surveillance companies, after being kept in the dark about the agency spending for years due to an earlier agreement allowing the department to bypass standard procurement rules for surveillance materials.

Shea’s subsequent disclosures made no mention of the NYPD’s murky relationship with Clearview AI — a company that scrapes billions of public photographs from social media sites without permission and distributes the information to police officers across the country.

Former NYPD Commissioner Raymond Kelly sits on the company’s advisory board. The police department has no formal agreement with the firm, but Legal Aid obtained emails documenting the relationship. Fox Cahn believes officers continue to utilize the software through individual trial accounts — something the NYPD strongly denied.

The department said it is constantly updating its guidelines as new technologies emerge, and changed its facial recognition rules to ban the use of Clearview AI in 2020.

The NYPD’s software is regularly reviewed by the National Institute of Standards and Technology to assess its accuracy and compare demographic data. Matches cannot be used as the sole basis for an arrest, and its policies are now clearly posted for the public to comment on, department spokesperson Jessica McRorie, a police sergeant, said in an email.

“In response to studies around bias, we recognize that there are concerns,” McRorie said. “This is why facial recognition is used in combination with human analysis, as NYPD human investigators are integrated into all the phases of the process and facial recognition findings are reviewed by these investigators.”

New York City Public Advocate Jumaane Williams, who is running for governor in the June Democratic primary, voiced support for some of the NYPD’s high-tech tools, like ShotSpotter. But he firmly opposes the use of facial recognition, citing concerns over flawed algorithms and what he views as undue privacy violations. He is part of Amnesty International’s Ban the Scan coalition, which seeks to outlaw the use of the technology in New York City.

“There is always a willingness to give up civil liberties when it is to protect yourself from ‘the other,’” he said. “But we have to strike a balance. And especially with the inherent flaws and concerns about data, I think [facial recognition] tips that balance.”