Claude is not a private diary. Anthropic handed over a threat from a conversation to the police

She perceived her conversation with Claude as her “diary”. However, a threat she wrote in it reached the police via Anthropic’s human reviewers. A woman from Lee County, Florida, now faces charges of making a written or electronic threat of mass shooting or terrorist act, Tom’s Hardware reports. The case demonstrates a specific boundary of privacy when conversing with AI: the content of the chat can be read by someone other than its author.

From Chatbot Message to Human Review

According to the police report described in the article, the user identified as Carli threatened to shoot up the Lee County Sheriff’s Office. The following day, she allegedly mentioned a new weapon in the conversation. Police subsequently apprehended her at home without incident.

The described process involves an intermediate step that is not visible during typical chatbot interaction. The police report states that Anthropic monitors chats for keywords and potentially threatening content. Depending on the severity, the conversation may then be escalated for human review. It was this team that decided to forward the findings to the police in this particular case.

What Privacy Rules Allow

Anthropic has not commented on the case itself, according to the article. However, its public privacy policy accounts for the disclosure of data to law enforcement: the company may do so if it believes in good faith that it is reasonably necessary to prevent serious harm to a person or property.

An interesting detail lies in the statistics. Anthropic’s report on government requests for the second half of 2025 lists zero emergency requests from the police. However, the company’s own notifications are not included in this category. Thus, the zero figure does not indicate how many times Anthropic itself alerted the police to dangerous content.

For using AI as a personal diary, this implies a significant difference compared to the idea of a private conversation. A confidential tone of responses is no guarantee that the content will remain solely between the user and the chatbot.

Reporting to the Police May Not Result in Charges

Tom’s Hardware also recalls an August case from San Francisco. A user in a chat with Claude threatened Anthropic CEO Dario Amodei and claimed to have purchased an AR-15 rifle. The company contacted the police, but the user was ultimately not arrested or charged. Thus, the mere handover of a conversation does not determine the outcome of a case.

In the Florida case, it is currently an indictment, not a verdict of guilt. Further details may emerge from the court hearing scheduled for November 2.

Do you perceive conversations with AI as a private diary, or as communication with an online service?

Sources: tomshardware.com