He lived 300 miles from the McDonald's. A 93% match sent him to jail anyway.

The Night at the McDonald's

Just before midnight on November 2, 2023, an adult man walked into a McDonald's in Jacksonville Beach and approached a girl under 12 at least twice, asking her to leave with him. She refused. On his second approach, she called for her mother. The man left before police arrived.^[1][4]

A Jacksonville Beach police officer, Scott O'Connell, took the case. According to the complaint later filed by the ACLU, O'Connell did not pull a copy of the McDonald's surveillance footage. He took cell phone photos of the surveillance screen, with the suspect's face "low resolution," "partially shadowed," and "off-axis."^[2][4] That is the image that eventually got fed into a facial recognition system and produced a 93% match to a man named Robert Dillon, who lives 300 miles away in Fort Myers, at the other end of the state.^[1][4]

By every account in the public record, the suspect was not Dillon. The McDonald's manager told investigators the suspect was a "regular customer" she had seen at the restaurant before.^[1][4] The order he placed to pick up had been paid for in advance through the McDonald's mobile app, which means McDonald's has the customer's name, payment details, and contact information. Jacksonville Beach police never asked McDonald's for any of that.^[2] The suspect had used a real, traceable identity to order the food, and no one followed that thread.

Where the 93% Came From

With no suspect and no other leads, O'Connell sent the cell phone photos of the surveillance screen to surrounding agencies in an "attempt-to-identify" bulletin. A sergeant with the Jacksonville Sheriff's Office (JSO) ran the photos through FACES, the Face Analysis Comparison and Examination System, a database operated by the Pinellas County Sheriff's Office since 2001.^[1][3][4] FACES runs Idemia algorithms and, at its peak in 2021, held tens of millions of Florida mug shots and driver's license photos, accessible to 263 local police departments, the FBI, and ICE.^[1][4]

FACES returned a ranked list of possible matches. The top hit was Robert Dillon. The score the system emits "represent[s] how much two images look alike to the algorithm. Not how likely it is that they show the same person," WIRED reported.^[1] The 93% the system spat out was a similarity score, not a probability of identity, and a 2016 Georgetown Law study of FACES found the Pinellas County Sheriff's Office "conducted no audits of how the database was searched and required no reasonable suspicion to run a query."^[1] When Georgetown's researchers asked Pinellas Sheriff Bob Gualtieri whether his office audited searches for misuse, he answered: "No, not really."^[1]

Two things should have stopped the case cold. They didn't.

First, a McDonald's employee picked Dillon's photo out of a lineup.^[3][4] A lineup identification of a person a witness has never seen before, generated by a system that itself produced the candidate, is not an independent confirmation. It is the same face, presented twice. The complaint flags this exact failure mode: "When facial recognition technology generates a false match, it will often be to someone who looks similar to the suspect, the lawsuit contends. That could mislead witnesses, who are asked to choose between that innocent look-alike and a set of random filler photos."^[4]

Second, O'Connell ran Dillon's name through a license plate reader database covering the two vehicles registered to him, looking for hits in the 48 hours surrounding the attempted abduction. Neither vehicle turned up in the Jacksonville Beach area.^[1][2] The complaint alleges those negative results were omitted from the warrant application O'Connell eventually submitted.^[1][2]

Nine Months of Silence

November 2023 turned into December, then January. Six months passed with no further investigation.^[1] In July 2024, O'Connell submitted the warrant. A judge signed it. In August 2024, deputies arrested Dillon at his home in Fort Myers, in front of his wife.^[1][2][3][4]

Stone crab season is when a commercial crabber makes most of his money. Dillon's arrest came at the worst possible moment. He posted bond by pledging the title to his truck, hired a defense attorney, and fell behind on rent. His mug shot stayed online on the county website for nearly a year. It was only removed after a TV reporter intervened.^[1][2]

When Dillon's defense attorney, Eric Rice, did what the Jacksonville Beach police apparently never bothered to do, he pulled Dillon's cell phone records and his work records. Dillon was at work on the day of the attempted abduction. He had been at work the day before and the day after. The State Attorney's Office dropped the charges within weeks of his October 2024 not-guilty plea.^[1][2][3]

The investigating officer, O'Connell, was promoted by the end of 2024.^[1]

Dillon v. City of Jacksonville Beach

On June 10, 2026, the ACLU, the ACLU of Florida, and the New York firm Hoguet Newman Regal & Kenney, LLP filed suit in U.S. District Court for the Middle District of Florida. The defendants are O'Connell and the JSO sergeant individually, plus the City of Jacksonville Beach, Jacksonville Sheriff T.K. Waters, and Pinellas County Sheriff Bob Gualtieri in their official capacities.^{[1][2][3][4][5]}

The complaint asks for compensatory and punitive damages, and asks a court to order all three agencies to overhaul their facial recognition policies, including a probable-cause floor that requires evidence beyond the algorithm's similarity score.^[1][3][4][5]

"One wrongful arrest is one too many. This should have never happened to Mr. Dillon," said Nicholas Warren, a staff attorney at the ACLU of Florida.^[5] "Florida's growing reliance on facial recognition technology threatens us all. We must stop this dangerous pattern before it traps more innocent people. No one should have their freedom taken away because the police rely on faulty technology."^[5]

"Robert's case illustrates the stakes when police deploy AI-assisted identification tools without adequate safeguards," said Steve Silverberg, counsel at Hoguet Newman Regal & Kenney. "Digital information can be a powerful tool for law enforcement, but its proliferation, supercharged by the AI boom, carries profound Fourth Amendment implications."^[4][5]

Jacksonville Sheriff T.K. Waters, one of the named defendants, had already told a local news station after the criminal case was dropped that a facial recognition match alone would not constitute probable cause in his office. "If you came to me with a facial-recognition hit and that was your probable cause, I would probably kick you out of my office," he said.^[1] The Pinellas County Sheriff's Office did not respond to WIRED's request for comment. Jacksonville Beach police told JaxToday they would not comment on the advice of their attorney.^[4]

The Florida Pattern

Dillon is not the first Florida case. He is the second this year, in the same sheriff's office, against the same Pinellas-operated system.

Earlier in 2026, the ACLU reported, the Jacksonville Sheriff's Office wrongfully arrested a North Carolina man in an auto-theft case after an 85% facial recognition match. The man spent nearly three months in jail. By the time the charges were dropped, he had lost his home, his job, and custody of his two children.^[1][5] Action News Jax covered that case. The Pinellas County system was the same one used on Dillon.^[1][5]

The pattern runs deeper than one agency. The ACLU's June 10 press release on the Dillon lawsuit explicitly flagged that "police in Maryland, Michigan, Missouri, Louisiana, Nevada, New Jersey, New York, North Dakota, Florida, and Arizona are publicly known to have wrongfully arrested people due to reliance on this technology."^[5] The ACLU's deputy director of the Speech, Privacy, and Technology Project, Nate Wessler, told WIRED: "No one should lose their freedom or be scared to leave their house because an algorithm got it wrong. These Florida police departments owe it to Mr. Dillon to make amends and to take serious steps to make sure this doesn't happen to anyone else. Police across the country are on notice. Unreliable face-recognition technology is hurting people, and we will keep fighting to hold them accountable for these abuses."^[1]

Reason's count puts Dillon at the 14th publicly known wrongful arrest from facial recognition.^[2] WIRED and Biometric Update put the count at 15, counting the North Carolina man as a separate case.^[1][3] The discrepancy is small and is the kind of detail a pattern-based litigation strategy tries to nail down: every false match in the country is, in principle, discoverable, and the cases the ACLU doesn't know about are the ones the organization cannot warn the public about.

What "93% Match" Actually Means

The wire story out of Jacksonville Beach says "93% match on facial features" and lets the reader fill in the rest. The rest is the entire problem.

Facial recognition similarity scores are not the probability that two images show the same person. They are a measure of how much two images look alike to a specific algorithm, on a specific image pair, with no built-in calibration for probe image quality. A higher-resolution, well-lit, on-axis probe image produces a more reliable score than a cell phone photo of a surveillance screen. In Dillon's case, the probe image was a phone snapshot of a TV screen displaying low-resolution McDonald's surveillance video, with the suspect's face partially in shadow. The ACLU's complaint cites the system's own technical assessment: facial recognition accuracy "depends significantly on the quality of the probe image. Lower-quality images contain less interpretable facial data, degrading the system's ability to produce a reliable template."^[2]

The 2016 Georgetown Law report on FACES flagged the same problem at the structural level: no reasonable suspicion required to run a query, no audit trail of who ran what search, no oversight body. A 93% match from a system with no internal audit and a garbage probe image is not evidence. It is a number.

Florida police departments have also used FACES to scan peaceful protesters, according to reporting by the Sun Sentinel and the Pulitzer Center.^[1] The same system, the same audit posture, the same kind of score.

The Bottom Line

Robert Dillon spent the night of August 2024 in a holding cell for a crime he did not commit, in a city he had never visited, on the strength of a 93% match produced by a 25-year-old facial recognition system with no audits, no reasonable-suspicion floor, and no requirement that the probe image be high quality. Jacksonville Beach police omitted the license plate reader result that would have exonerated him. The officer who built the case was promoted at the end of the year. Dillon's mug shot sat on a county website for nearly a year. The man who actually tried to lure the child at the McDonald's has not been charged, and may never be, because the case file points to the wrong person.

On June 10, 2026, the ACLU asked a federal court to make all three agencies answer for the pattern, not just the incident. "Florida police must implement safeguards and ensure this never happens to anyone else," Dillon said in a statement released with the lawsuit, "because until they do, nobody is safe."^[4][5]

The pattern is now ten states wide. The litigation strategy is now pattern-based, not incident-based. The 93% threshold is now the line the ACLU is going to push on. The 14th, or 15th, known wrongful arrest from a facial recognition match is now a federal civil rights lawsuit, and the next one will be too.

Sources

1. WIRED: "Wrongful Arrest Exposes Failures in One of the Oldest Police Face-Recognition Tools in the US" (Dell Cameron, June 10, 2026)
2. Reason: "ACLU Sues After Facial Recognition Falsely Identifies Florida Man as Child Abductor" (Joe Lancaster, June 11, 2026)
3. Biometric Update: "Wrongful arrest based on false FRT match sparks lawsuit from Florida man" (Joel R. McConvey, June 11, 2026)
4. Jacksonville Today: "Faulty facial recognition leads to lawsuit after Jacksonville Beach arrest" (Dan Scanlan, June 10, 2026)
5. ACLU: "Florida Man Sues Police Over Wrongful Arrest Due to False Facial Recognition Match" (June 10, 2026, primary press release)

Related Coverage

• A Grandmother Lost 5 Months to a Facial Recognition Error The Tennessee grandmother jailed for five months on a Fargo, North Dakota warrant after a Clearview AI match. Same pattern, different state, different defendant. The canary case for the Florida pattern that just produced Dillon.
• A Maryland Woman Spent Six Months in Jail on a Facial Recognition Error The Maryland case. Same template: low-quality probe image, no independent verification, months of jail before exoneration. The ACLU's pattern-based strategy started here.
• The 13 Dismissed Facial Recognition Cases Our running count of known U.S. wrongful arrests from facial recognition. The Dillon and North Carolina cases have now pushed the count to 14 or 15, depending on how the ACLU classifies the second JSO arrest.
• Retailers Are Filling Their Stores With Face Recognition. Most Shoppers Have No Idea. The commercial use of the same underlying technology, with the same audit and disclosure problems. The civil litigation strategy the ACLU is building for police is going to land on retailers next.
• EFF: "Move Fast and Surveil Things" The civil liberties case for why "ship it, then fix the false matches" is the wrong order of operations for any facial recognition deployment.