Robert Dillon's lawsuit against Jacksonville Beach, JSO, and Pinellas County turns a single wrongful arrest into a structural blueprint for what an algorithmic probable cause check should require.
The license-plate reader had already cleared him. So had the phone call. So, in any functional sense, had basic geography. Robert Dillon lives in Fort Myers, more than 300 miles from the Jacksonville Beach McDonald's where, on the night of November 2, 2023, a girl under 12 said a stranger tried to lure her out of the restaurant. An automatic license-plate search of Dillon's two vehicles for November 1 through 3, 2023 returned zero hits in Duval County. When Corporal Scott O'Connell of the Jacksonville Sheriff's Office called him before any warrant was sought, Dillon denied involvement, said he had never been to Jacksonville Beach, and described a distinctive scar running from his hairline to his nose.
None of it mattered. What mattered, according to the federal civil rights complaint Dillon filed this week in US District Court for the Middle District of Florida, was a 93% similarity score returned by FACES, the Face Analysis Comparison and Examination System maintained by the Pinellas County Sheriff's Office. The "image" fed into FACES was not a surveillance still. It was a photograph of a McDonald's computer monitor displaying surveillance video, taken by a detective's phone, per the complaint. A score of 93% does not mean 93% certainty of identity. It means the algorithm judged the probe image to be more similar to Dillon's mugshot than to 93% of the candidates in the database, with whatever error rate the deployment's threshold implies.
As Ars Technica reported on the filing, the affidavit that supported the warrant did not just rely on the score. It allegedly omitted the empty ALPR pull, omitted the phone call in which Dillon named the scar, and misrepresented a McDonald's manager as a witness to the attempted luring when, the complaint says, she was busy with work duties and had not seen the incident. The State Attorney dropped the charge roughly two months after Dillon's August 2024 arrest at his Fort Myers home, in front of his wife, after he had spent a night in jail and borrowed against his truck title to post bond.
The ACLU, which is representing Dillon alongside Hoguet Newman Regal and Kenley, framed the case in its press release on the filing as one of structural failure rather than individual malfeasance. The advocacy group called Dillon "one of 15 known people" in the United States to be arrested after a false facial-recognition match. That figure is an ACLU tally drawn from civil cases and news reports; the number of people actually charged after a false match is not centrally tracked, and the 15 is best read as a floor, not a count.
The technical chain matters as much as the score. FACES is administered by the Pinellas County Sheriff's Office. The Jacksonville Sheriff's Office has access to it, and Jacksonville Beach PD is among the partner agencies that can request searches through JSO. Sergeant James Walters, named as a defendant alongside O'Connell, JSO Sheriff T.K. Waters, Pinellas County Sheriff Bob Gualtieri, and the City of Jacksonville Beach, allegedly ran or supervised the search. The lawsuit's Monell claim targets the city for the practice, not the individual officer: that the FR use policy allowed a single algorithmic score, on a low-quality image, to substitute for the geographic, biometric, and witness cross-checks an investigator would otherwise have run.
The complaint also presses the qualified-immunity question harder than most FR cases. It alleges that O'Connell was previously terminated from the St. Johns County Sheriff's Office for threatening to "blow up" the agency, was reinstated, and was later arrested for domestic battery before resigning. If the court credits that history, the argument goes, no reasonable officer with that record could have treated a 93% score on a photo of a screen as probable cause without first checking the obvious exonerating evidence that was already in the case file. The history is asserted in the pleading and not yet adjudicated; it is included here because the complaint itself puts it at the center of the Monell and Brady theories.
The constructive question is what would have stopped this. The factual record suggests a short list of minimum evidentiary cross-checks before an algorithmic ID can ground a warrant. Investigators should have run a geographic check against Dillon's known address and recent movements, including ALPR hits and, where lawful, cell-site data, and put the negative result in the affidavit. They should have done a biometric cross-check, including the scar Dillon described in the pre-warrant call, matched against the underlying image rather than against the mugshot derived from it. The warrant affidavit should have disclosed that the probe image was a photograph of a screen displaying surveillance video, and at what resolution, so a judge could weigh what 93% means against that input. There should have been a lineup or witness procedure that did not depend on the algorithm, even informally, to test whether a human identifier reached the same conclusion. And the agency should have documented the threshold it relied on, including its own false-positive rate at the score it used, ideally cross-referenced against NIST FRVT demographic data and attached or summarized for the issuing judge.
Each of those steps would have surfaced an inconsistency in Dillon's case. The ALPR pull already had. The scar already had. The 300 miles already had. The cost of those checks is measured in investigator hours, not in new technology. The cost of skipping them, in this case, was a 52-year-old man booked on a charge of attempting to lure a child, in front of his family, for an incident the State Attorney ultimately did not believe had enough support to prosecute.
The court Dillon is asking to convict is not the one that dropped the charge. It is the one that signs warrants. What the complaint is really testing is whether "the algorithm said 93%" can stand in for probable cause when the same case file already contains the evidence that would have unraveled the match. The answer the City of Jacksonville Beach, JSO, and Pinellas County file in response will determine whether FACES stays a hint to investigators or becomes, in practice, the investigation.