Rogue NYPD officers are using a sketchy facial-recognition software that its own facial recognition unit doesn’t want to touch because of concerns about security and potential for abuse, The Post has learned.
The NYPD’s facial-recognition unit kicked the tires on a controversial app Clearview, which has scraped millions of photos from social media to help ID suspects — the department, though, ultimately passed on using it, with one department insider likening the program to “playing with fire.”
But that hasn’t stopped several dozen of cops from outside the department’s facial recognition unit from using the app “to this day” — with the company claiming the last photo search was registered at 10:56 a.m. Thursday.
As part of an ambitious nationwide sales pitch to both federal and local law-enforcement agencies, Clearview AI presented its tech to the NYPD in early 2019, according to police sources.
The department’s facial recognition unit took Clearview up on a complimentary 90-day trial, but restricted its use to non-investigatory testing, not field police work, because Clearview could not say who had access to images once they were loaded into its massive database, sources said.
By snapping a picture of someone and uploading it into the app, a user is provided with public-posted photos of that person in Clearview’s database, culled from millions of Web sites including Facebook, LinkedIn and even mobile-pay system Venmo.
More than 600 law-enforcement agencies have embraced the app in the past year, from local to federal agencies, including Gainesville, Fla. Police Department, the FBI, US Homeland Security and Indiana State Police, the company told The Post.
On Wednesday, Twitter issued a cease-and-desist order to the Clearview, telling it to stop crawling the site for photos, a company spokesman said.
“We are reviewing it,” the spokesman said, claiming the terms and conditions were not violated and their program had been vetted by attorneys.
Although the NYPD decided not to integrate the tech into its facial recognition unit, it did not deny that individual cops have taken advantage of the free trial.
“While the NYPD looked at it, the facial indication section never used it for investigations,” a police source said noting that its standard practice for the department to review any new tech or equipment.
Department insiders said, however, that they walked away troubled by the implications of the technology.
One fear, a source said, was the potential to abuse the system for extracurricular searches.
“It only takes one cop to put in his ex-girlfriend’s photo in there and see who she’s dating now,” said the insider. “They’re playing with fire. It’s going to catch up to them.”
A spokesman for Clearview claims, “Numerous investigators from around the department are using the app to this day” — with about 36 active accounts using the program for months logging thousands of searches. The company would not provide exact numbers or which case was searched Thursday morning, citing the internal specific numbers as confidential.
In addition to the facial-recognition unit, Clearview pitched the app to other units like veritable door-to-door salesmen, a source said.
Also concerning to the NYPD, sources said, were the ties of Hoan Ton-That, the app’s creator, to viddyho.com, a site involved in a widespread phishing scam that plagued Gmail users in 2009, as reported at the time by Gawker.
Though the NYPD ultimately turned down Clearview, the company continued to insist on — or at least strongly imply — a link.
The start-up’s Web site prominently features a video compilation of police investigations aided by facial-recognition technology, including the NYPD’s bust of a kook who sparked a bomb-scare by leaving two rice cookers in a Downtown Brooklyn subway station in August.
A Clearview internal executive summary obtained by The Post similarly suggests that their tech helped the NYPD identify the suspect, Larry Griffin II.
“The images below show real examples of Clearview’s matching algorithm,” the report reads, above a side-by-side of two photos of Griffin, one a surveillance still from the bomb-scare and one a mugshot from an unrelated arrest.
“All results were confirmed as accurate matches through subsequent investigation,” the summary adds. “All examples are from real law enforcement investigations.”
Clearview says an agent from a New York-based federal agency, which it declined to name, linked Larry Griffin to a West Virginia.
The only problem, the NYPD said, is that Clearview’s app had nothing to do with making the connection.
“There is no institutional relationship between the NYPD and Clearview,” said Kaye. “The NYPD identified the suspect using the Department’s facial recognition practice where a still image from a surveillance video was compared to a pool of lawfully possessed arrest photos.
“A facial recognition match is merely a lead,” Kaye added. “No one has ever been arrested solely on the basis of a computer match, no matter how compelling.”
Source: Read Full Article