A 30-year-old Bonita Springs, Florida woman was arrested in September after writing explicit threats against the Lee County Sheriff's Office in what she described to law enforcement as her private AI chatbot "diary" — unaware that the platform's automated safety monitoring system had flagged her messages, that human reviewers had assessed them as credible threats, and that the company had notified law enforcement without requiring a subpoena.
According to the arrest report cited by Gulf Coast Now, the woman had an ongoing conversation with Anthropic's Claude AI assistant in which she wrote a series of threatening statements directed at the Lee County Sheriff's Office. The first threat, timestamped September 26 at 5:10 a.m., stated: "I'm going to shoot up the sheriff's right the [expletive] now." A follow-up message sent in the early hours of September 27 stated: "This is 100% last chance I'm done. I got a new [expletive] gun today. [Expletive] you." The messages were flagged by Anthropic's automated safety monitoring system, which routes messages meeting certain risk thresholds to human reviewers. Those reviewers assessed the messages as sufficiently serious that they warranted emergency disclosure to law enforcement — an exception to Anthropic's standard policy of not sharing user information without legal process. Deputies arrested the woman at her residence without incident. She was charged with written threats of violence, a second-degree felony in Florida carrying a potential sentence of up to 15 years in prison and a $10,000 fine.
"She uses Claude like a 'diary.'" — Lee County Sheriff Carmine Marceno, relaying the explanation the woman offered to law enforcement after her arrest
The legal framework Anthropic operated under is straightforward. The company's published policies state it does not disclose user information in response to government requests without valid legal process — but carves out an emergency exception when it believes there is an imminent risk of physical harm or death and providing information without delay may avert that emergency. The woman's messages, appearing in the pre-dawn hours and explicitly mentioning a new firearm, evidently crossed that threshold in the judgment of the human reviewers who assessed them. The disclosure did not require a warrant, a subpoena, or any legal process at all. Florida law on written threats of violence does not require evidence of intent to carry out the threat — the written threat itself, if serious and unconditional, constitutes the offense.
The case is a significant public reminder of something that should not require a reminder but apparently does: conversations with AI chatbots are not private in the way a journal or a conversation with a therapist might be. They are logged by technology companies, subject to safety monitoring, and — under specific circumstances involving imminent safety risks — can be disclosed to law enforcement without the user's knowledge or consent and without requiring any legal process whatsoever. Anthropic's emergency disclosure policy is appropriate and reasonable. It exists because the alternative — treating all statements made to an AI as confidential regardless of content — would be indefensible if a preventable act of violence occurred after the company had been explicitly warned.
Written threats of violence under Florida Statute 836.10 are a second-degree felony when the threat is directed at a person, a group, or a place and is communicated in writing — including electronic communications. The statute does not require proof of intent to carry out the threat. Several courts have held that AI chatbot conversations constitute written electronic communications within the scope of similar statutes in other states. The woman's identity was not publicly released by Lee County authorities as of publication.
People have a right to express frustration, anger, and even violent ideation — in appropriate contexts, with appropriate people, who understand the difference between venting and planning. An AI chatbot is not a therapist, not a confessor, and not a diary. It is a software service run by a corporation that has both legal obligations and genuine safety responsibilities. Anyone using AI conversation tools for emotional processing would be well served to understand that distinction before typing something into a chatbox that they would not want read aloud in a courtroom.