A Florida woman was arrested after writing a shooting threat in a private Claude chat that was flagged by Anthropic’s safety system, according to the report carried by BlockBeats. The conversation was then sent to human review, and the review team passed the material to police. She is now facing felony charges.
Local police said she later explained that she used AI as a "diary." An arrest report said she first wrote that she wanted to attack the Lee County Sheriff’s Office, and the next day said she had just gotten a new gun.
Police contact was decided by humans, not Claude
The report said Claude did not decide on its own to alert law enforcement. The system first identified a potential threat, then routed serious content to Anthropic’s human team, which decided whether to contact police.
Anthropic’s consumer terms state that the company may, at its own discretion, report a user’s inputs, outputs, or actions to law enforcement. Its privacy policy also allows the company to share personal data with police when it believes doing so is necessary to prevent serious harm to people or property.
Even when a user has disabled model training, conversations flagged for safety reasons may still be reviewed by the safety team.
Legal questions remain
The case is also facing legal debate. Florida law requires that this kind of written or electronic threat be sent or transmitted in a way that can be viewed by another person. Local lawyers said applying that statute to a private message written to an AI is not simple.
For now, the woman has only been charged and has not been convicted.
Not an isolated case
This is not the only such incident. Tom's Hardware counted at least three cases since August in which Anthropic chat content came to the attention of police.
In an earlier case, a Claude user threatened Anthropic CEO Dario Amodei in a conversation, after which Anthropic contacted San Francisco police.