When Aurora police wanted to know the identity of two armed men protesting outside a police station, they sent a screenshot from a video to the Colorado Department of Motor Vehicles and asked investigators there to run the image through facial recognition software.
Investigators with the DMV were able to match one of the men against the agency’s database of driver’s license photos and detectives identified him. The other man couldn’t be identified.
The case in Aurora is one of hundreds of times Colorado law enforcement in the past few years have accessed facial recognition software through the DMV or by using a program more than 80 Colorado agencies paid for that has become increasingly common across the state.
Few of the agencies that adopted the software made public announcements about what investigators say is a key instrument in fighting crime.
“Facial recognition is an incredible tool that helps us identify a lot of serious suspects out there,” Lakewood police Cmdr. John Pickard said.
But civil rights leaders and tech experts say there are serious concerns about the use of such technology, especially when the public doesn’t know it’s being used. Research has shown some facial recognition programs are prone to error when images of people with darker skin are submitted. Critics worry that without proper oversight and strict regulation, such programs could result in an invasion of privacy or, worse, the arrest of an innocent person.
Colorado has no laws regarding facial recognition software, though law enforcement officials say they are willing to work with policy makers on guardrails to the programs.
“As always, our technology has outstripped our laws and our morals,” said Steve Beaty, a computer science professor at Metropolitan State University of Denver.
Driver’s license comparisons
Aurora police investigators asked the DMV to use its facial recognition software on the men at the protest because officers inside the surrounded station feared they could be shot if they tried to leave, according to the arrest affidavit for one of the protest organizers arrested Sept. 17. Police believe the men were acting as security to the crowd of several hundred.
“Those two people who were armed caused a lot of fear and concern and we have to give a good effort to identify who that person is,” Aurora police spokeswoman Faith Goodrich said.
But even with the successful identification, neither of the men were arrested in connection with the July 3 protest in Aurora, Goodrich said.
Aurora police’s query to the DMV about the armed men is one of dozens such requests the department receives every year from law enforcement. The DMV added facial recognition software to its programs in 2011, department spokeswoman Julie Brooks said.
Data from the Department of Motor Vehicles shows that since July 2016, local and state law enforcement agencies in Colorado have made 227 requests that the department use its facial recognition to help in an investigation. Federal agencies made 94 requests in that time, as did two out-of-state agencies. The data does not say which local and state agencies filed those requests and a DMV spokeswoman said the agency does not track that information.
Since July 2016, the DMV has denied 22 of 323 requests from law enforcement.
The DMV’s policy states that it will only complete facial recognition comparisons for outside law enforcement for alleged felonies as well as identity theft and fraud cases that the department’s investigative unit is also working. The policy says comparisons for other types of cases can also be approved by a supervisor on a case-by-case basis.
Data from the department shows the comparisons were requested for a wide range of alleged crimes. Some of the most common reasons listed were homicide, passport fraud and sex crimes against children. Other searches were given more vague reasons, including “intelligence” and 55 cases listed as “unknown.”
In three-quarters of cases, the DMV was unable to locate photos that matched images provided by outside law enforcement agencies. The department’s policy specifies that only “high probable” matches be provided to law enforcement.
The cooperation between the DMV and law enforcement is permitted under state statutes that say criminal justice agencies can access DMV records in connection to the agencies’ “official duties and functions.” There are no memorandums of understanding or agreements between agencies about the information sharing, Brooks said.
Denise Maes, public policy director at the ACLU of Colorado, said she didn’t know that this information sharing existed and she was surprised to learn of it.
“Why is there this cozy relationship between law enforcement and the DMV — I don’t think there ought to be,” Maes said.
One of the problems with such a program is that many people posing for a driver’s license photo don’t know that image will be used in such a program and compared to suspected criminals, Beaty said. If there is a case of mistaken identity, it could cause police to start looking into the life of a person who isn’t connected to any crime.
“We need to ask ourselves: How surveilled do we want to be as a society?” he said.
But the driver’s license photo database is not the only way Colorado police have used facial recognition software.
Growing use of facial recognition
Up until July, Lakewood police officers could snap a picture of a suspect on the street, enter the photo into an app on their phones and use software that would compare the submitted photo to a database of mug shots collected by a group of Colorado law enforcement agencies.
It was an useful tool, Pickard, the police commander, said. It helped police officers identify people who refused to give their real identity, though officers are trained to find other evidence to confirm any identity matches suggested by the software.
“The one thing that we preach is that you never identify anyone with facial recognition, it is only an investigative lead,” Pickard said. “If you do it wrong it can lead to problems.”
In July, however, Lakewood police and the 84 other Colorado law enforcement agencies that had access to the software lost facial recognition capabilities when the vendor, LexisNexis, disabled that part of one of its programs, called Lumen.
“We made the decision to disable the image matching feature until legislation and policies outlining usage of the tool are in place,” LexisNexis spokeswoman Sara Herrmann said in a statement. “The image matching feature will resume once clear guidelines on the use of facial recognition for law enforcement are in place.”
Colorado law enforcement agencies have been using Lumen since 2016 to collect and share data on arrestees and wanted persons. The program added facial recognition capabilities in 2017.
All 84 agencies participating in the Colorado Information Sharing Consortium had access to Lumen and about 80% of those agencies used it on a regular basis, said David Shipley, executive director of the consortium.
The Lumen program compared a photo submitted by an investigator — including those from security cameras, social media profiles, dating websites and photos taken on scene — to the database of more than 8 million jail booking photos compiled by the consortium.
The photos in the database are not removed if a person is acquitted or charges are dropped unless that person successfully petitions to have those records expunged or sealed, Shipley said.
That’s concerning, said Beaty, the Metro State professor, because it ignores a person’s right to be presumed innocent until proven guilty and also keeps people in the database even after they’ve served their sentence.
“I think we need to ask the question, what do we mean by someone who paid their debt to society?” he said.
Investigators, patrol deputies and crime analysts at the Arapahoe County Sheriff’s Office started accessing Lumen’s facial recognition in 2018, Capt. Jared Rowlison said. The agency didn’t track how often deputies were using the software or how many cases the software helped solve, but Rowlison said it has been crucial in cracking many, particularly thefts.
One case involved a woman whose date attempted to sexually assault her. She met the man through a dating website and didn’t have much information beyond a first name and a photo after he deleted his profile. Investigators were able to run that photo through the Lumen system and narrow down possible suspects before eventually arresting the man.
Testimonies about the facial recognition program posted on the Colorado Information Sharing Consortium’s website state the software has been used to help solve frauds, hit-and-run cases, armed robberies and sex crimes against children.
Opting out of new tech
Some agencies, like the Denver Police Department, have opted not to use facial recognition. An effort to ask voters to put a ban on facial recognition in the city failed to collect enough signatures to put the question on the ballot, though other municipalities like San Francisco have enacted such restrictions.
“The ACLU has been adamantly opposing facial recognition on the grounds that it’s a threat to privacy when it does work and it’s a racist threat to everyone when it doesn’t work,” Maes said.
Studies by academics as well as the federal government have shown that facial recognition software is less accurate when trying to identify women or someone with darker skin.
Although the Arapahoe County Sheriff’s Office has no particular policy about the use of facial recognition, Rowlison said his deputies are bound by the same rules that limit their use of other law enforcement databases. Any possible identifications suggested by facial recognition need to be confirmed through more investigation, he said.
“We take those concerns to heart and still use the information at our hands appropriately in pursuit of an investigation,” he said.
The public should be informed any time a law enforcement agency is using a new surveillance tool, especially facial recognition, Maes said. Agencies should be required to appear before city councils or other governing bodies to explain any new surveillance tool they’re using, she said.
“The only way to get a handle on it is to let the public know,” she said.
Neither Lakewood police nor the Arapahoe County Sheriff’s Office made any notification to the public when they started using the technology.
“We’ve had plenty of people ask us about it and we’re very open to the conversation and how we’ve used it,” Rowlison said. “It’s not something we’ve tried to hide.”
Both Pickard and Rowlison said they’ve explored other facial recognition software vendors while Lumen is disabled.
“Really, I’m just biding my time until LexisNexis allows us to have access again,” Rowlison said.
This content was originally published here.