A United States judge has delivered one of the most consequential rulings yet on the use of artificial intelligence in law enforcement, holding that the Fourth Amendment does not permit police to arrest a person based on nothing more than an uncorroborated AI conclusion. The decision, issued in a civil suit brought by Alain Vizcay Osorio and published via DocumentCloud, rejects a detective's attempt to claim qualified immunity after he arrested Osorio for allegedly passing a bad check based solely on a facial recognition match — a match the court found was obviously wrong. The ruling's opening paragraph states the principle flatly: "The Fourth Amendment to the United States Constitution does not allow a detective to delegate to AI the task of assessing probable cause." For ongoing coverage of AI policy and regulation, AI Buzz Wire is tracking the appeal that many expect to follow.

A Bad Check, a CCTV Screenshot, and a Facebook Match

The case began with low-stakes alleged crime and low-effort police work. Osorio was accused of passing a bad check at a local bank, and the only images of the suspect came from the bank's CCTV footage. Detective Ryan Kilmer of the Florence, Kentucky Police Department took a screenshot of the suspect's face and ran it through the department's facial recognition program. The system returned a match to a Facebook profile belonging to someone listed as "Yony Smith Npoles Charon" — and, for what the court dryly called an "unknown reason," Kilmer concluded that the profile belonged to Osorio.

The photographs told a different story. According to the decision, the CCTV footage and the Facebook profile image "do not resemble the booking photo of Osorio." Based on the booking photo, the court observed, the only thing Osorio had in common with the suspect was that both were Black men with facial hair; the photos showed a different head shape, a different haircut, different facial features, and a different complexion.

Kilmer's affidavit told the court he had used "investigative tools" to "match" Osorio to the suspect. In reality, the judge found, he used a single tool and accepted its faulty output — despite Florence PD policy explicitly stating that facial recognition alone is not enough to establish probable cause. The detective, the ruling makes clear, ignored both his department's written rules and the Constitution's baseline requirement that probable cause reflect an actual human assessment of the evidence.

What the Court Actually Held

The decision's core language goes beyond a finding that this detective was negligent. It articulates a categorical rule: "the Fourth Amendment does not permit an individual to be arrested based on nothing more than an uncorroborated conclusion by AI that the individual in question has committed a crime. And no reasonable detective in the Defendant's position could conclude otherwise."

That second sentence does the heavy lifting. By stating that no reasonable officer could believe an uncorroborated AI match suffices, the court demolished the foundation of the qualified immunity defense, which generally shields officials from civil liability unless they violated clearly established rights. Kilmer, the court held, cannot hide behind the doctrine — both because the Fourth Amendment itself forbids what he did and because his own department's policy told him, in writing, that facial recognition alone cannot establish probable cause.

The practical consequence is that the detective faces the lawsuit personally, a posture that both increases the pressure to settle and makes appellate review more likely, since the officer now has a direct financial stake in challenging the ruling.

Implications That Reach Far Beyond One Bad Check

Legal commentators note that the logic of the decision radiates outward across the fast-growing market for police AI. The same reasoning — that an algorithm's output is not evidence of probable cause until a human independently corroborates it — applies with equal force to:

  • Facial recognition searches, including systems that match surveillance stills against scraped web images, the pattern at issue in this case.
  • Predictive policing software that directs patrols and public interactions based on algorithmic forecasts.
  • AI-generated police reports, now being added to body cameras by vendors like Axon on the premise that officers prefer not to write them — raising the question of whether a machine-drafted narrative of a stop or use of force can constitute the official record.

If a court says an algorithm cannot make the probable-cause determination, every downstream system that quietly assumes the algorithm did make that determination inherits the problem. Departments that have treated AI outputs as investigative shortcuts may need to document genuine human corroboration before arrest, or risk both exclusion of evidence and personal liability for the officers involved.

What Comes Next

The immediate question is appeal. Because qualified immunity was denied, the detective has strong motivation to take the case upward, and the constitutional question — whether and how AI tools can participate in probable-cause assessments — is exactly the kind of issue appellate courts would need to settle for a rule this broad to harden into precedent. Civil liberties groups are likely to intervene in support of the ruling; police unions and AI vendors are likely to push back, arguing that the decision conflates a sloppy investigation with a technology policy problem.

Either way, the ruling sets a marker that police departments can read today, without waiting for the appellate cycle: AI output is a lead, not a conclusion. The detective in this case let a face-matching algorithm do the thinking, ignored his department's own rule against exactly that, and now faces the consequences personally. As the decision itself signals, the Constitution's framers did not anticipate neural networks — but the requirement that a human being assess probable cause before the state seizes a person turns out to translate without any amendment required.

For departments shopping for AI tools, the message is equally direct: vendor claims about accuracy are not policy. Written rules requiring human corroboration — like Florence PD's, which the detective ignored — are now not just best practice but the difference between a department that survives a lawsuit and an officer who pays for it personally.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →