Robert Dillon’s phone buzzed on the table one August morning in Fort Myers. He answered, expecting another order for blue crabs, maybe a wholesale pickup from Naples. Instead, he heard the voice of authority: "Robert Dillon? You’re under arrest for attempting to lure a child." The deputy stood at his front door, body cam rolling. His wife watched from the porch, still in her robe.
He was 52 then. A commercial crabber for more than three decades, who knew tides better than most people know their own birthdays. He’d never been to Jacksonville Beach—over 300 miles away—and he hadn’t left Lee County in nearly two years, not since his daughter started high school. Yet the deputy gave him three seconds to fetch shoes before cuffing him in front of their pickup truck.
Two months later, the state dropped the charges. His record was wiped. But his mugshot? That’s still up somewhere on a county database, accessible to anyone with an internet connection. Neighbors still ask about the case in the grocery aisle. He flinches when children near him at the truck stop.
This wasn’t a case of bad luck. It was a cascade failure: low-quality surveillance, an overreliant AI, and an officer who chose to ignore every fact that didn’t fit his story.
Dillon is now suing the City of Jacksonville Beach, the Jacksonville Sheriff’s Office (JSO), and the Pinellas County Sheriff’s Office for damages. His complaint paints a chillingly familiar script: algorithm first, investigation never.
The truth? The man who police claim tried to lure a child at a Jacksonville Beach McDonald’s never stood at that counter. And the evidence that would’ve proven it sat—literally—in an officer’s warrant affidavit, quietly omitted.
A pixelated ghost, and a 93% certainty that wasn’t real
The whole thing started on November 2, 2023. Just before midnight at a Jacksonville Beach McDonald’s, a manager called police. A man, the story went, had asked a girl—under twelve—twice if she wanted to leave with him. She said no both times, phoned her parents next door at their hotel, and the suspect walked out.
The police got the surveillance video from inside the restaurant. That’s where things started to unravel.
The source footage itself was low-grade, recorded on a store camera that no one bothering to calibrate or clean. Then someone photographed the screen of a desktop monitor playing that footage back, capturing glare, distortions, and a second layer of pixelation. This wasn’t HD footage handed to a forensic analyst; it was the equivalent of taking your phone and snapping a picture of your TV screen during a storm.
The Jacksonville Beach Police forwarded the still images to JSO, which runs the Face Analysis Comparison and Examination System (FACES), a centralized database with over 38.5 million images—including mugshots and driver’s license photos—from 196 agencies across Florida.
The system returned Robert Dillon as a 93% match.
Here’s the part people forget: that number means nothing in plain English. According to Dillon’s lawsuit, "a 93% match" isn’t a probability of identification. It’s just how close two mathematical representations are—like comparing pixel patterns, not faces. An officer handed that percentage to a magistrate and said, "I need a warrant." The rest—the alibi, the contradictory evidence—can wait.
The ACLU put it differently in its press release: “He is one of 15 known people to have this happen to them in the United States.” Fifteen. In broad daylight, across a nation of hundreds of millions.
The officer knew. And he chose to pretend he didn’t.
Corporal Scott O’Connell was the lead investigator. He’d already been fired—then reinstated—by another Florida sheriff’s office after threatening to "blow up" his agency. Later, he was arrested on domestic battery charges and resigned before those cases could proceed.
Jacksonville Beach hired him anyway. And in August 2024, they made him corporal.
This is where the lawsuit really stings: not because O’Connell made a mistake, but because he never tried to get it right.
Here’s what he omitted from his arrest warrant affidavit:
-
LPR evidence. A search of the license plate reader database confirmed neither of Dillon’s vehicles was in Duval County on November 1–3, 2023. If he hadn’t driven to Jacksonville Beach—and no other rentals show up—he couldn’t have committed the crime.
-
Dillon’s own words. O’Connell called Dillon weeks before applying for the warrant. Dillon denied involvement, stated he’d never been to Jacksonville Beach, and described a distinctive scar from his hairline to his nose. O’Connell didn’t mention that call once.
-
The department’s own policy. JSO’s guidance says facial recognition results "cannot constitute a positive identification, are inherently unreliable, and do not constitute probable cause." Yet O’Connell presented the 93% score as though it were evidence, not a dead end to be investigated.
-
The McDonald’s manager. The affidavit implied the manager had witnessed the incident firsthand. In fact, she was too busy at work to pay close attention—and had described the suspect as a "regular customer." That detail never made it into the warrant, either.
The complaint doesn’t mince words: “O’Connell did not merely fail to investigate. He instead affirmatively chose not to pursue readily available investigative avenues.” What he ignored:
- Mobile ordering and payment records from the McDonald’s app
- Digital receipts, timestamps, or account history tied to that night
- older surveillance footage showing the regular’s prior visits
- a side-by-side of Dillon’s scar against the suspect’s face in the footage
- cell phone location pings from the time and place of the alleged crime
The only thing he pursued with vigor was confirming his own suspicion.
The photo lineup wasn’t a test. It was a trap.
Here’s how the misidentification snowballed:
After FACES flagged Dillon, O’Connell requested a photo array—essentially a lineup—but the fillers were chosen to resemble Dillon, not the suspect in the surveillance photo. That meant Dillon stood out as the one who looked most like the person in the grainy image—not because he matched, but because everyone else looked less like him.
The ACLU explains this well in its complaint: when facial recognition gives a false positive, it often lands on someone who shares features with the real culprit. Then a lineup with “fillers” chosen to look like that innocent person makes the mismatched face feel… right. It’s psychological priming, wrapped in procedure.
To top it off, the McDonald’s manager viewed Dillon’s photo array—but the child victim never did. No forensic psychologist. No neutral facilitator. Just an officer who got a match he liked and moved on.
The system worked exactly as designed: confirmation bias baked into algorithmic output, handed to a witness who’d already been told who the suspect was.
Dillon told Gulf Coast News later: “Says it’s 93 percent accurate. Far as I’m concerned, it’s 100 percent inaccurate.” He pointed out the suspect’s hair: "black wavy." Dillon’s isn’t. The algorithm saw something. He wasn’t it.
One arrest. Two wrongful cases in the same system.
Dillon was held overnight, then released on bond. He had to borrow money and pledge the title to his truck just to walk out of jail. His income—driven by crabbing’s seasonal rush—dried up for nearly a month. He couldn’t focus on work. Couldn’t go anywhere without someone recognizing his face from the local news.
"No law enforcement agency has ever apologized or acknowledged the error," his lawsuit says.
But here’s what makes Dillon’s case even more alarming: he wasn’t alone.
The ACLU press release points out a second wrongful arrest in Florida—just last week—by the same Jacksonville Sheriff’s Office. A North Carolina man spent three months in jail after JSO relied on a FACES match. He lost his job, custody of his kids, and half a year of his life.
And that’s just what made the headlines.
FACES has been in use since before facial recognition entered the public consciousness. But this isn’t a tech failure. It’s a process failure: one where an error-prone tool, left unchecked, becomes authority. Where the algorithm’s answer replaces the officer’s judgment—and where that judgment isn’t just flawed, but actively concealed.
The suit asks for damages and policy reforms. But what Dillon really wants, he told reporters: "No one should lose their freedom or be scared to leave their house because an algorithm got it wrong."
A 93% certainty isn’t proof. It’s just the end of a shortcut—and in this case, someone paid the price for taking it.