Edward Longe is director of technology and innovation and director of national strategy at the James Madison Institute in Tallahassee.
In August 2024, police arrested Robert Dillon at his home in Fort Myers, Florida, claiming that he had attempted to abduct a child at a McDonald’s in Jacksonville Beach, roughly 300 miles away. Dillon insisted that he had never been to that city.
The case rested on an artificial intelligence facial recognition search that had matched Dillon’s face to grainy surveillance footage at the restaurant. Police used this information, along with a restaurant employee’s statement, to arrest Dillon.
Prosecutors ultimately dropped the charges, but the damage had largely been done. Dillon had been publicly accused of being a predator and was forced to hire a lawyer to prove he had been elsewhere at the time of the alleged incident. In the United States, the state is supposed to prove guilt. In this case, Dillon had to prove his innocence.
Dillon’s case is not an isolated one. Earlier this year, Florida prosecutors dropped charges against Jalil Richardson, a North Carolina man accused of selling a stolen car in Jacksonville. Facial recognition software had flagged him as an 85 percent match to the seller, and two witnesses had then identified him. A judge said that was enough for probable cause. Richardson spent nearly three months in jail before his attorney produced time cards showing that he was working in North Carolina at the time. By then, he had lost his job, his home and custody of two of his children.
According to the ACLU, since 2019, at least 14 people nationwide have been wrongfully arrested after police relied on a faulty facial recognition match. While the technology is clearly highly useful to law enforcement, safeguards against its misuse must be put into place.
In both cases described above, the technology worked as intended. The facial recognition systems that law enforcement used searched databases and found matches that looked like the suspects. The problems began when law enforcement treated that response as evidence, rather than a lead.
Witnesses then selected matches from a photo lineup. But if facial recognition systems have already identified two individuals who resemble the suspects, it’s highly probable that humans will do the same, turning the photo lineup into a rubber stamp for AI facial recognition.
Jacksonville prosecutors claimed the match and the two eyewitness identifications were enough for probable cause in Richardson’s case. In Dillon’s case, according to the lawsuit he has filed against the city of Jacksonville Beach and multiple sheriff’s offices, a license plate reader search showed no sign of his car near the restaurant. The evidence that could have cleared him was there but was ignored.
Defenders of facial recognition technology will inevitably say these cases are technical glitches, but they point to broader violations of the personal liberty that the U.S. justice system is supposed to protect. The Fourth Amendment to the U.S. Constitution has typically required police to show probable cause for an arrest, and that standard is supposed to consider the totality of the circumstances and rest on facts about a particular person. A similarity score only measures how much two faces look alike.
The frightening part is that almost anyone can end up on the wrong side of these systems. A decade ago, researchers at Georgetown Law estimated that more than 117 million American adults were already in police facial recognition networks. They also found that no states require officers to have reasonable suspicion before running a search, and that for the most part, searches weren’t audited for misuse. Most Americans have never agreed to be part of a searchable lineup, but many of them are. Any of them could be accused of a crime they never committed, in a town they’ve never visited.
None of this means that police should abandon facial recognition technology. Used properly, it’s a valuable tool for law enforcement. After the 2018 shooting at the Capital Gazette newsroom in Annapolis, Maryland, police used it to confirm the identity of the gunman, who had refused to cooperate. The Miami-Dade Police Department is using facial recognition to identify human-trafficking victims and break up human-trafficking rings. An outright ban on the use of the technology would remove a valuable tool only because some departments have used it carelessly.
Police can keep the tool without sacrificing anyone’s liberty. In 2024, the city of Detroit reached a settlement with one Robert Williams that prohibited police from making arrests based solely on information acquired from facial recognition technology. The settlement also required independent evidence to corroborate any match. In 2024, Maryland likewise passed a law stating that a facial recognition match cannot be the sole basis for probable cause.
Other cities and states should follow the lead of Detroit and Maryland. Police should treat a match as a lead but never as probable cause. They should be required to obtain independent evidence before presenting a match to a witness, and judges and defendants should be told whether the technology has been used in a case.
Facial recognition will only get more powerful, and as it does, more police departments will use it. That’s why preserving the standard matters. An algorithm can tell police whose face looks similar to a suspected perpetrator’s, but it cannot tell them who committed a crime.
Figuring that out is still the job of the people to whom we entrust the power to arrest.
The post Let’s face it: This valuable technology is easy for police to misuse appeared first on Washington Post.
