Robert Dillon sues Florida cops: a “93% match” facial ID triggered a wrongful arrest

Robert Dillon says Florida police arrested him after a facial recognition system flagged him as a “93 percent match” to a suspect captured on McDonald's surveillance video. He claims the match was wrong and that he had never set foot in Jacksonville Beach, even though officers allegedly used that AI score to move from suspicion to prosecution.

According to the lawsuit filed today, Dillon was targeted after police ran a low-quality image against a facial recognition system. The image, the suit says, was a photo taken of a McDonald's computer screen that was displaying video surveillance footage. Dillon lives more than 300 miles from Jacksonville Beach, and the lawsuit says a police search of a license plate reader database found no evidence he was in the area at the time of the alleged crime. The charge: attempting to lure or entice a child under twelve years old, stemming from an August 2024 incident at a Jacksonville Beach McDonald’s.

This is the kind of allegation that makes executives in security, risk, and compliance sit up. In high-stakes policing, the “AI did it” story can turn into a procedural shortcut: if an algorithm returns a confident percentage, investigators may treat that output like a substitute for the unglamorous work of verifying evidence. Dillon’s lawsuit portrays that shortcut as the core failure, arguing that officers did not test the machine’s answer against evidence that would have cleared him. Instead, he alleges, they built a case to confirm the system’s result.

The lawsuit also raises an issue that matters beyond one man. Facial recognition systems are often advertised as fast, scalable, and useful for investigative leads. But when they intersect with degraded or indirect data, the risk profile changes. Here, the suit points to a particularly problematic input: the flagged photo was reportedly taken from a McDonald's computer screen showing surveillance footage, which is a reminder that “garbage in, garbage out” is not just a tech slogan. The lower the quality and the more indirect the capture, the more room there is for a bad match to look plausible.

There is a second-order effect too: once law enforcement builds a narrative around a technology output, it can become harder to back out without admitting error. Dillon alleges officers concealed exculpatory evidence, and the suit frames the case as what happens when police let an error-prone artificial intelligence system stand in for an investigation. In operational terms, that alleged concealment is a force multiplier. It can lock in confirmation bias, because the evidence that undermines the AI match never gets to do its job.

For decision-makers who advise organizations using or deploying facial recognition, this is a governance and controls story as much as a civil-liberties story. Boards and senior leaders typically think about model performance metrics, false positives, and data provenance in product contexts. The courtroom version is different: it asks what humans did with the output, what standards were required before acting, and whether exculpatory information was handled transparently. Dillon’s allegations put the spotlight on the human-in-the-loop question, not just the model.

There is also the regulatory and public-trust backdrop. Across the country, facial recognition has faced mounting scrutiny, with regulators and advocacy groups pushing for limits, audits, and stronger safeguards. While this specific lawsuit is about Dillon’s case in Florida, it fits into a wider pattern: the legal system is gradually forcing organizations to treat biometric matching as a consequential decision, not an optional lead. When someone is arrested based on an AI match that is later disputed, the stakes are not abstract. Dillon argues his arrest and prosecution involved one of the most stigmatizing crimes a person can face. That kind of reputational and personal harm is precisely why governance frameworks matter.

Finally, the case has strategic implications for peers across policing, public tech, and enterprise security. If an “accuracy percentage” becomes a de facto credibility badge, institutions may underinvest in verification steps that are uncomfortable but necessary. Dillon’s lawsuit alleges officers should have tested the machine’s answer against evidence that would have cleared him, including distance and location checks like the license plate reader database results. For leaders, the takeaway is blunt: confidence scores can mislead, especially with low-quality inputs. The real question is whether procedures require independent corroboration before an algorithmic match turns into an arrest.