What the arrest report says
On 26 September a user in Bonita Springs, Florida wrote into Anthropic’s Claude that she was going to shoot up the Lee County Sheriff’s Office. The following day, according to the arrest report, the same account said she had got a new gun.
Anthropic’s platform monitors for key phrases and potentially threatening content. Investigators wrote that because of the severity of the statements, the material was escalated to a human review team, which then passed it to law enforcement. Deputies identified the woman, went to her home and detained her without incident; an intelligence detective at the Lee County Sheriff’s Office took over the case.
She is charged with making a written threat of violence under Florida law, a felony, and is due in court in November. WINK News reported the arrest on 30 September; the case was picked up more widely on 5 October. Sheriff Carmine Marceno said she later told investigators she used the chatbot like a diary.
The policy that allows it
None of this required a warrant or a court order. Anthropic’s published usage policy prohibits using its products to facilitate or promote violence or intimidation, and its policy on government requests says user information may be disclosed to law enforcement in limited emergencies where there is a danger of death or serious physical injury. That is the clause the company appears to have acted under. WINK News asked Anthropic for comment on how its safeguards handle threats and when it shares information; the report does not record a reply.
Marceno used the case to make a point about anonymity, telling WINK News that people using AI chat and search are never truly anonymous. A retired FBI agent quoted in the same report said courts are still working out how to handle threats involving AI.
The mechanism is not new — major platforms have reported threats to police for years, and US law already requires providers to report some categories of material. What is new is the surface. A chatbot invites a kind of writing people do not do in a public post: unfiltered, addressed to no one, closer to a private notebook than to a timeline.
What is not settled
Two things about this case will come up again. The first is the gap between how private a chatbot feels and how private it is. A professor at Florida Gulf Coast University, quoted by WINK News, said people turn to AI partly because a machine does not judge them, and put the warning bluntly: “AI is not your best friend.”
The second is the standard. Anthropic’s safeguards flagged the phrases; a human team judged them credible; the sheriff’s office agreed. At each step someone made a call about intent from text alone, without tone, context or the person in front of them. Florida’s written-threat statute applies to communications sent in a manner in which another person may view them, and whether a private chat with a model meets that test is exactly the kind of question a court has not answered.
The November hearing is the first place it might be.
