A woman in Florida used Claude like a private diary. One entry ended up with the police. On 26 September she wrote that she would shoot up the Lee County Sheriff’s Office, and a later message mentioned getting a new firearm. Anthropic’s systems flagged the conversation, a human reviewer escalated it and the company reported it to law enforcement, Tom’s Hardware reported.
Four days later she was detained at her home in Bonita Springs without incident. She now faces a felony charge under a Florida law against written threats of violence.
How the message reached police
Anthropic watches conversations for phrases that could signal a threat. Most are handled automatically, but the most serious are escalated to people for review, and that is what happened here. The reviewers decided the risk was real enough to pass the details to law enforcement.
Florida treats a written or electronic threat to carry out a mass shooting or act of terrorism as a second degree felony. Deputies identified the woman, visited her home and detained her before a detective took over the investigation.
Anthropic has not commented publicly on this case. Its approach mirrors how email providers and social networks handle credible threats: automated systems narrow down a huge volume of messages, and people make the final call on what goes to the authorities.
What Anthropic’s privacy policy allows
This is not a hidden rule. Anthropic’s privacy policy says it may disclose personal data to government authorities or law enforcement when it believes in good faith that this is reasonably necessary to prevent serious harm to any person or property, or to comply with the law.
The policy also says conversations can be flagged for safety review to detect harmful content and enforce its rules. That applies even to people who have opted out of having their chats used to train models.

Not the first chat to reach police
It is at least the third AI conversation to reach police since August, according to reporting on the case. Earlier examples include a threat against a San Antonio elementary school on 11 August and a threat against a chief executive in San Francisco on 14 August. A separate case involving ChatGPT came to light in March.
Each case follows the same pattern. People treat a chatbot like a private notebook, but it is a service run by a company with safety teams and legal obligations.
Why Australian businesses should care
The lesson for employers is not about threats. It is about what staff assume is private. Many people paste customer details, health information and internal disputes into AI tools as if they were talking to themselves, when the provider can review flagged content and disclose it in some circumstances. Business plans often come with different data terms from personal accounts, so it also matters which version staff are actually logged into.
Your AI policy should spell out which tools staff may use, on which plans, and what must never go into them. Our Privacy Act AI playbook sets out what the Privacy Act 1988 expects when personal information goes into AI systems.
Private, but not secret
Few people will argue with stopping a credible threat of violence. The broader point is that AI chats sit somewhere between a diary and an email to a company, and users should treat them that way.
For businesses choosing a platform, the safety review and disclosure terms deserve the same attention as price and features, something we look at in every AI consulting engagement.
