What She Told an AI Chatbot Led to Her Arrest

Person using smartphone and laptop with AI hologram — Photo: MMD Creative / Shutterstock

The uncomfortable truth at the center of this case is not that an algorithm read someone’s words — it’s that millions of people now compose their most private thoughts inside a product that was never, legally or technically, a diary at all, and the law has not caught up to that confusion.

Key Points

A Lee County, Florida woman, Carli Michelle Heller, was arrested on a second-degree felony charge after an alleged September 26 Claude conversation in which she said she would “shoot up” the Lee County Sheriff’s Office
Anthropic’s automated system flagged the language, a human Trust and Safety reviewer judged it credible, and the company referred the conversation to law enforcement under its own privacy policy, not a court order
Florida Statute 836.10 criminalizes transmitting a written or electronic threat “in a manner in which it may be viewed by another person” — a standard written for emails and social posts, not private chatbot exchanges
No warrant or subpoena preceded the referral, and critics argue the case exposes an unresolved gap between a company’s “sole discretion” to report users and a user’s reasonable expectation of privacy
The arrest establishes probable cause, not guilt — intent, transmission, and the statute’s applicability to AI conversations remain open legal questions likely headed for a courtroom test

What The Arrest Record Actually Shows

Strip away the headlines and the documented spine of this case is straightforward. According to the redacted arrest affidavit filed under case number 26CF017549, Detective Alejo found probable cause to charge Heller with violating Florida’s written-threats statute after a September 26 conversation with Claude in which she allegedly stated she would “shoot up” the Lee County Sheriff’s Office. A follow-up message the next day reportedly referenced acquiring a new firearm and described a “last chance” — context that moved the exchange from an ambiguous outburst toward something investigators treated as a specific, target-identified threat. That second detail matters: isolated venting reads very differently from venting paired with a weapon reference and a named government target.

Anthropic’s own account, as described in the affidavit, is that Claude’s system monitors conversations for key phrases and threatening content, and that messages crossing a severity threshold are routed to a human Trust and Safety review team before any outside disclosure occurs. That two-step structure — automated flag, human judgment — is worth underscoring, because it rebuts the simplest and most viral version of this story: that a chatbot unilaterally called the police. It did not. A person reviewed the material and made the referral. Whether that person’s judgment was sound, however, is precisely where the case becomes contested rather than settled.

The Mechanism Behind The Referral

Anthropic’s privacy policy, in force since September 10, 2026, reserves the right to disclose user data when the company believes in good faith that doing so is reasonably necessary to prevent serious harm, alongside standard provisions for valid legal process. Critics characterize this as a “sole discretion” clause — language that lets the company decide, on its own authority, when a conversation crosses from private to reportable. This is not unique to Anthropic. A federal statute, 18 U.S.C. 2702(b)(8), already permits providers to disclose user communications to law enforcement when they believe in good faith that an emergency involving danger of death or serious injury requires it without delay. OpenAI has described a comparable practice, routing chats that suggest an “imminent and credible risk of harm” to specialized review and, in some instances, to the FBI, with reporting describing at least one resulting conviction. Heller’s case is not an aberration — it is the visible edge of a policy architecture that most major AI platforms already have in place.

What makes this case sharper than most is the statute prosecutors chose. Florida’s written-threats law requires that a threat be transmitted “in a manner in which it may be viewed by another person” — language drafted for emails, texts, and social posts meant for an audience. Attorney and commentator Carne Shagal Feforn, discussing the case publicly, raised exactly this tension: users widely believe a chatbot exchange is private, and the law’s “transmission” and “viewability” language was never written with a machine intermediary in mind. Whether a conversation a user believed was confidential satisfies a statute built around visibility to other people is not a settled question — it is the question this prosecution will have to answer.

Where The Genuine Dispute Lies

The factual record of what Heller allegedly wrote is not seriously contested; the arrest affidavit and multiple outlets converge on the same quoted language. The real disagreement is legal and structural. Commentary following the arrest points out, accurately, that no warrant or subpoena preceded Anthropic’s disclosure to police — the company acted under its own contractual authority, not under judicial compulsion. That raises a live Fourth Amendment question Feforn flagged publicly: does a user retain a reasonable expectation of privacy in a chatbot conversation that a company’s staff can access and hand to police at its own discretion? Separately, First Amendment “true threats” doctrine, most recently clarified by the Supreme Court in Counterman v. Colorado, holds that prosecutors generally must show a speaker was at least reckless about whether their words would be perceived as threatening a specific person — intent and context matter, not just the words on the page. Whether a frustrated diary-style entry, never intended for an audience, meets that bar is exactly the kind of fact-intensive question a jury or appellate court, not a headline, will eventually resolve.

A Florida woman told deputies she uses Claude like a diary. Police say one entry, saying she would "shoot up" the Lee County Sheriff's Office, got her arrested on a felony charge.

Here's what the reporting says. The arrest report says Anthropic's system flagged the message, a… pic.twitter.com/XRETRtdoNd

— Laura Hartley (@LauraHartleyCO) October 7, 2026

What This Means Going Forward

The practical lesson for anyone using a commercial AI chatbot is unglamorous but important: these systems are not confidential in any legal sense comparable to a locked journal, a doctor’s office, or attorney-client communication. They are company-operated products, subject to terms of service that explicitly reserve the right to monitor, retain, and in defined circumstances disclose what users type. The Heller case will likely serve as an early test of how far that authority extends when the statute being enforced assumes a human audience and the user assumed none at all. Until courts resolve that gap, the responsible assumption — for users and for the companies building these tools — is that anything typed into an AI chat can be read by more than the AI.