# Osorio v. Kilmer: Court Says Police Cannot Hand Probable Cause to AI

> Source: <https://explainx.ai/blog/osorio-v-kilmer-ai-facial-recognition-probable-cause-qualified-immunity-denied-2026?utm_source=rss&utm_medium=feed&utm_campaign=blog>
> Published: 2026-11-20 00:00:00+00:00

A federal court in Kentucky has refused to let a detective escape a wrongful-arrest lawsuit built on an AI facial recognition match. In *Osorio v. Kilmer*, decided November 9, 2026 in the Eastern District of Kentucky, the court wrote that the Fourth Amendment "does not allow a detective to delegate to AI the task of assessing probable cause," according to [a write-up of the opinion](https://ohiogovernmentactually.com/posts/signs-point-to-yes/?ref=explainx). [Techdirt](https://www.techdirt.com/2026/11/19/court-tells-cops-that-the-4th-amendment-doesnt-cover-ai-generated-probable-cause/?ref=explainx) covered it on November 19 under a headline that reads the opposite way round.

The headline overshoots in another way too. This is a ruling on a motion, not a trial. The court assumed Osorio's account was true and said discovery could change the picture. This post sets out what happened, what the court held, what it left open, and what the Hacker News thread added.

## Quick reference

| Question | Answer | 
|---|---|
| Case | *Osorio v. Kilmer* , 2026 WL 3390238 (E.D. Ky. Nov. 9, 2026) | 
| Who is suing whom? | A Nebraska man (Osorio) is suing Florence, Kentucky detective Ryan Kilmer under Section 1983 | 
| What did the AI do? | Matched a bank-camera image to a booking photo and a Facebook profile under another name | 
| What did the court decide? | Denied qualified immunity for now; claims go on | 
| Final liability finding? | None. Allegations are taken as true at this stage | 
| Tool named? | Not in the sources we read | 
| Binding elsewhere? | No: a district court opinion is persuasive only | 

## What happened to Osorio

A dome surveillance camera with a green indicator, the kind of image source behind a facial recognition probable cause dispute

The facts below are the plaintiff's allegations as the court accepted them, per the [opinion summary](https://ohiogovernmentactually.com/posts/signs-point-to-yes/?ref=explainx).

Someone tried to cash a suspected fraudulent check at a Kentucky bank using a fake Texas driver's license. Detective Kilmer ran a security-camera image through facial recognition. It returned a booking photo of Osorio and a Facebook profile under a different name.

Kilmer's warrant affidavit said he used "investigative tools" and that the photos matched. Osorio alleges they did not match, and that the affidavit left out that software had made the identification. He was arrested in Florida, extradited to Kentucky and spent over two weeks in jail before he posted bond. The indictment was later dismissed with prejudice after he showed he was working in Nebraska when the check was presented.

One more allegation matters a lot. Osorio says the detective's own department policy described facial recognition as "advisory" and said it did not establish probable cause without further investigation.

## What the court actually held

Kilmer asked for judgment on the pleadings on the Section 1983 claims. The court declined, and three points from the summary matter most.

1. **A possibly misleading affidavit.** The allegations were enough to suggest a misleading warrant affidavit. Without the challenged statements, what remained showed that someone committed a crime, not that Osorio was that person.
2. **Qualified immunity denied at this stage.** Qualified immunity shields officers unless they violated clearly established law. The court said Osorio did not need an earlier case about AI facial recognition. Sixth Circuit precedent on unreliable evidence and recklessly false warrant statements was enough.
3. **A "Magic 8 Ball" backstop.** The court added a hypothetical: an officer who arrests on a Magic 8 Ball would obviously violate the Fourth Amendment, and AI used as a sole basis is no different. That ties to the "obvious case" exception to qualified immunity.

The court did not rule on municipal liability. It also said Kilmer may still establish qualified immunity after discovery.

So the accurate summary is narrower than either headline. The court did not ban facial recognition and did not find that a match can never help. It held that, taking the allegations as true, a detective cannot treat an AI output as a stand-in for his own assessment, and cannot leave out that the identification came from software.

## How probable cause, warrants and qualified immunity fit together

Three legal ideas explain why this ruling reads the way it does. None of them is specific to AI, and that is the point.

**Probable cause** is the Fourth Amendment standard for an arrest or a search warrant: enough facts that a reasonable person would believe this particular person committed the crime. Knowing that a crime happened is not enough. The facts must point at the person.

**A warrant affidavit** is the sworn statement an officer gives a judge to get that warrant. If an officer recklessly includes false statements, or leaves out facts that would change the judge's view, a later court can strike those parts and ask whether anything is left. In Osorio's case, the court said that without the challenged statements, the remaining content showed only that someone committed a crime, not that Osorio was that person.

**Qualified immunity** protects officers from personal liability unless they broke clearly established law. Plaintiffs normally have to point to an earlier case with similar facts. There is a narrow exception for obvious violations. The Magic 8 Ball line is the court explaining why an obvious-case argument is available even though no earlier AI case exists.

Put together, the logic is simple. An officer who swears that photos "match" and omits that software made the call may be treated as having misled the judge. An AI tool changes how the lead was produced. It does not change who is responsible for the sworn statement.

## Why this lands now: a run of facial recognition cases

This is not the first court to look at the question. Earlier cases pointed the same way, and the pattern is what makes this ruling worth watching.

- In Ohio, the trial court in *State v. Tolbert* treated a Clearview AI match like an anonymous tip that cannot alone establish probable cause, per[ACLU of Ohio's amicus coverage](https://www.acluohio.org/press-releases/aclu-ohio-files-amicus-brief-raising-concerns-about-use-facial-recognition-technology/?ref=explainx) and[BiometricUpdate](https://www.biometricupdate.com/202501/judge-throws-out-clearview-facial-recognition-match-in-us-murder-case?ref=explainx) .
- In July 2026 a judge in Washington, D.C. ordered disclosure of Clearview AI's role in a facial recognition arrest, per [BiometricUpdate](https://www.biometricupdate.com/202607/judge-orders-disclosure-of-clearview-ai-role-in-dc-facial-recognition-arrest?ref=explainx) .

Our earlier coverage of the infrastructure behind this problem is in [Flock cameras and AI surveillance](https://explainx.ai/blog/flock-cameras-ai-surveillance-civil-liberties-2026) and the [federal bill to cut surveillance funding](https://explainx.ai/blog/tlaib-get-the-flock-out-of-our-communities-act-federal-surveillance-funding-ban-2026). Those posts cover how matches get produced. This one covers what happens when a match is wrong.

## What developers are saying

The [Hacker News thread](https://news.ycombinator.com/item?id=50567735&ref=explainx) was small but active. Three comments are worth noting, all as the commenters' own readings, not verified facts.

- User AnimalMuppet [pointed out](https://news.ycombinator.com/item?id=50567883&ref=explainx) that this is not a suspect trying to block a warrant. The case was dismissed, and the former suspect is now suing the officer. They read the court's wording as meaning the officer cannot use qualified immunity, and called it a big deal.
- User jp57 [objected to the headline](https://news.ycombinator.com/item?id=50568016&ref=explainx) , arguing that "doesn't cover" is confusing and that the article means the Fourth Amendment disallows AI-generated probable cause.
- User Molitor5901 [objected](https://news.ycombinator.com/item?id=50568307&ref=explainx) to the article's suggestion about the detective's motives, saying it was an assumption and "not in evidence."

The first reading goes further than the opinion as summarized. The court denied immunity at the pleading stage and said Kilmer may still establish it later.

Agents that act on your behalf need rules. [See how explainx helps secure your agents →](https://explainx.ai/agentbeam?src=blog_note)

## Why it matters for people who build AI systems

If you build or buy systems that rank faces, plates or people, the case shows a failure mode that has little to do with model accuracy.

- **A match is a lead, not an identification.** A similarity score tells an investigator where to look. It cannot say that a person committed a crime.
- **Provenance has to travel.** The allegation that the affidavit omitted software involvement is the legal hook. Systems should record that an output came from a model, which model, and what a human checked next.
- **Policy beats product copy.** The department's own rule called the tool advisory. Courts notice when practice contradicts written policy.
- **The buyer is not the only defendant.** Vendors who market "identification" language invite the very over-reliance that ends in litigation.

Other stories show the same gap between an automated flag and a human decision: an [AI gun detector that flagged a Doritos bag at a Baltimore school](https://explainx.ai/blog/doritos-bag-ai-gun-alert-lawsuit-baltimore-schools-2026), and a report that the [IDF used Amazon Rekognition on millions of photos](https://explainx.ai/blog/idf-amazon-rekognition-3-5-million-photos-palestinians-guardian-2026). For background on why error rates differ across groups, see our guide to [bias in AI](https://explainx.ai/blog/what-is-bias-in-ai-types-examples-mitigation-2026).

## What people are asking

### Is a facial recognition match enough to arrest someone?

Not on its own, according to this court's reasoning, and not according to the detective's own department policy as alleged. Corroborating work is what turns a lead into probable cause. This is one district court, not a nationwide rule.

### Can I sue if AI misidentified me?

Wrongful-arrest claims under Section 1983 are the usual route, and *Osorio* shows one surviving a first challenge. Outcomes depend on the facts, the jurisdiction and qualified immunity. Talk to a civil rights lawyer. This post is not legal advice.

### Does this ban police use of facial recognition?

No. The sources describe a ruling on one arrest and one warrant affidavit. Nothing in them restricts using software to generate investigative leads.

### What should I watch next?

The docket. Discovery may show which tool was used, how the match was scored and what the detective checked. An appeal to the Sixth Circuit would matter more than the district court opinion.

## What is confirmed and what is not

**Confirmed from the sources:** the court, the date, the denial of qualified immunity at this stage, the quoted holding on delegating to AI, and the reported timeline of the arrest and dismissal.

**Not confirmed:** whether the faces actually matched (that is Osorio's allegation), which software was used, what discovery will show, and how the department policy will be read at trial. The details here come from a detailed written summary of the opinion and Techdirt's coverage, so check the docket for the full text before quoting it in anything formal.

## Related reading

*Court details are accurate as of November 20, 2026 and may change as the case moves through discovery.*
