The Detroit Police Department has dramatically reduced its reliance on facial recognition tools to help identify violent crime suspects following a legal settlement that strengthened rules for how the technology can be used.
DPD used facial recognition to identify possible suspects wanted in connection with violent crimes nine times last year, the department said, which is a 91% drop from 2023.
DPD told BridgeDetroit in a statement that the department’s use of the technology started to decrease in 2024, when it was used 28 times. The department said it doesn’t have data or statistics about the number of times facial recognition has led to an arrest or conviction.
Detroit police has defended its use of facial recognition software as a tool to produce investigative leads in violent crime and home invasion cases. Critics of the technology have long opposed the use of the technology to identify suspects due to its inability to accurately identify Black and Brown faces.
“If it’s not being used (or) hardly at all, that’s a good thing. It’s something we really want to reserve for the last resort,” Board of Police Commissioners Member for District 7 Victoria Camille told BridgeDetroit during a recent community policing event at the University of Michigan Detroit Center.
Since the department started using the technology in 2017, there have been at least three lawsuits against the city over misidentification. A 2024 settlement agreement in one of those cases has led to policy changes and procedures to prevent future instances of misidentification.
To use facial recognition, the police submit a “probe” photo, a video still of a suspect’s face, and the technology searches databases of mugshots and driver’s license photos to produce potential matches based on facial features and other physical characteristics.
Only one of the nine facial recognition searches Detroit police conducted last year produced an investigative lead, a BOPC report shows. Three searches were related to murders, three for aggravated assaults, and two for robberies. All of the requests were for Black males.
Of the 115 search requests submitted in 2020, 60% resulted in a possible match, a BOPC report shows. Aggravated assaults accounted for 67 requests, 37 for robberies and three for murders.
The remaining were related to arson, kidnapping, criminal sexual conduct, threats against police, damage to property and home invasion. All but four of the unidentified individuals were Black, and the majority were males.
Why DPD’s use of facial recognition technology decreased
Detroit police first purchased facial recognition software through South Carolina-based DataWorks Plus for $1 million in 2017. After using it without a policy for nearly two years and amid intense public criticism, former police chief James Craig proposed the department’s first policy governing its use.
The BOPC approved the first facial recognition policy in 2019 to protect against potential police abuse. Among the provisions:
- Facial recognition software could only be used in cases of suspected violent crimes or home invasions.
- Prohibited for surveillance through livestreams or on mobile devices.
- Prohibited for use at ‘First Amendment events’ such as protests and other constitutionally-protected activities.
- Police analysts must verify any facial recognition matches the software produces and get that verification approved by a supervisor.
- Any match is only a lead for investigators and can never be used as the sole basis for an arrest.
- Penalties were established for police misconduct that could include termination. DPD was also required to inform the BOPC and city officials of any violations.
- Required weekly reports to the BOPC on how the software was being used.
“We think those new policies represent best-in-the-nation approaches to the use, and restrictions on the use, of facial recognition technology, ” American Civil Liberties Union (ACLU) attorney Ramis Wadood told BridgeDetroit in an interview.
Even with the new rules, it didn’t prevent problems.
Detroit police continues to defend its use of the technology as a tool to produce investigative leads in violent crime and home invasion cases, even after three cases of misidentification and subsequent lawsuits against the city – Robert Williams and Michael Oliver in 2019, and Porcha Woodruff in February 2023.
Detroit police used facial recognition to identify Williams as a potential match for a suspect wanted in connection with a 2018 theft that occurred at Shinola downtown.
A Shinola employee was shown photos of six potential suspects and asked to identify the individual he saw on surveillance video, despite not being present when the theft occurred, and he picked Williams. The photo used to identify Williams was produced as a possible match using facial recognition technology, and was later proven to be incorrect.
Williams repeatedly denied being the individual in the photo when asked by detectives.
Under the settlement agreement in Williams’ case, the department was prohibited from arresting people based solely on facial recognition results and from conducting lineups based solely on facial recognition investigative leads without independent and reliable evidence linking a suspect to the crime — all of which was included in additional policy changes the department implemented in August 2024, and according to the department, ultimately led to the decrease in its use.
“We’re glad to see that those stricter policies are having the intended impact, where police are using facial recognition much, much less, and in the rare and special circumstances that the policies allow,” Wadood said. “We’re glad the number of searches dropped considerably since the settlement, but that number at the end of the day should be zero. It’s still a nascent technology that study after study has shown is simply not reliable, especially when it comes to darker skin colors.”
The settlement in the Williams case also called for an audit of all instances since 2017 in which the Detroit Police Department used face recognition technology to obtain an arrest warrant. A spokesperson for DPD confirmed Friday that the audit was completed, but wasn’t immediately able to provide BridgeDetroit with a summary of its findings.
The Detroit Police Department told BridgeDetroit it doesn’t currently spend any money on facial recognition, through DataWorks Plus or any other vendor, and hasn’t since 2022. Before then, the department said it spent $2,373,084.30 on the technology.
Michael Peterson II, administrator of Detroit’s Community Violence Intervention (CVI) program said reliance on the technology to solve crimes is not only lazy, but it’s also “unhealthy and harmful.”
“That’s not the end-all, be-all. Some of these tools are in place, and they are helpful, we can’t deny that technology has made things easier. But that is not enough,” Peterson told BridgeDetroit at the UM event.
Members of the community also serve as the eyes and ears of their neighborhood and can be a valuable resource for law enforcement in identifying a person involved in crime, according to Camille.
“If we have the good relationships in communities that we’re supposed to have, we’re all walking facial recognition,” Camille said. “We know who did it. We don’t need a machine to tell us who did it.”
Not just a Detroit issue
According to the ACLU, at least 14 people across the country are publicly known to have been wrongfully arrested by police because of reliance on erroneous facial recognition results.
When the ACLU sued Detroit police on behalf of Williams in 2021, supporters of police using facial recognition technology characterized his wrongful arrest as an unfortunate but isolated mistake that shouldn’t undermine trust in the technology. But the wrongful arrests have kept coming, the ACLU says.
For example, the ACLU cited, many of the people wrongfully arrested had visible differences compared to the suspect in the photo: Michael Oliver had full tattoo sleeves, Porcha Woodruff was eight months pregnant.
At least 15 states have enacted policies that restrict or regulate how the technology is used. New Jersey, for example, requires defendants to be notified when it’s used in an investigation, while other states, such as Maryland, require a warrant to use it.
“We were luckily able to fix a lot of problems in Detroit, but these problems are popping up in cities, towns, and counties across the state,” Wadood said.
Wadood told BridgeDetroit that although Detroit has received a lot of attention for its use and misuse of facial recognition, the issue extends to local police agencies across the state, especially since the Michigan State Police allows other agencies to use its facial recognition technology.
“Thankfully, the Detroit Police Department isn’t spending millions of taxpayer dollars on its own technology, but it still has taxpayer-funded technology through the state police,” Wadood said.
For example, if a local police agency wants to run a facial recognition search they can submit a request to the state police, who will then run a search and send the results back to the requesting agency.
Detroit Police Media Relations confirmed that it does have a Memorandum of Understanding with MSP for this purpose and includes those searches in its monthly reports to the BOPC.
MSP First Lt. Mike Shaw didn’t respond to questions about agreements with Detroit police or other local agencies, and instead, referred BridgeDetroit to its website for more information about the use of facial recognition technology.
According to MSP’s website, approved law enforcement agencies can request a facial recognition search through MSP’s Statewide Network of Agency Photos (SNAP) Unit, and a trained facial examiner will conduct a search on their behalf. Local agencies can also request access to MSP’s SNAP facial recognition desktop tool to conduct a search on their own, which only allows them to search mugshot and arrest databases.
Shaw also directed BridgeDetroit to file a Freedom of Information Act (FOIA) request to find out the number of times DPD requested a facial recognition search through the state police between 2020 and 2026, which ultimately came with the price tag of $1,000 to produce the records. BridgeDetroit narrowed its request and is awaiting a response.
MSP’s website says that banning law enforcement’s ability to use facial recognition would force facial examiners to manually analyze images, “resulting in lengthy, inefficient, and costly investigations,” and puts the public at a greater risk of victimization.
“DPD recognizes that facial recognition technology has its challenges, but we recognize the usefulness of the software. Facial recognition is another tool that may be used in investigations to assist with identifying a suspect,” DPD media relations told BridgeDetroit in an emailed statement.
