A conversation with Claude on September 26, 2026 became the direct lead that brought Lee County, Florida deputies to the home of a 30-year-old woman. According to the Lee County Sheriff's Office arrest report and local outlet WINK News, Carli Michelle Heller of Bonita Springs wrote in a conversation with Anthropic's chatbot Claude that she planned to attack the local sheriff's office. The content triggered Claude's safety systems, was escalated to human review, and a reviewer who judged it a credible threat reported it to law enforcement. Deputies detained Heller at her home; she told investigators she had been using Claude as a diary. She now faces a charge of making a written threat of violence. The case has not been tried in court, and she has not been convicted.
What happened between the automated flag and the call to police
According to police, Anthropic's safety mechanisms monitor keywords and potential threats, escalating content by severity. The decisive step in this case was not the algorithm but the emergency disclosure after human review: Anthropic's public policy says the company normally shares user information with governments only after valid legal process, but an exception exists where it believes an emergency could cause imminent physical injury or death and timely disclosure might prevent it. Florida Statute 836.10 makes it a potential second-degree felony to threaten, through a written or electronic record that another person may view, to kill or injure someone, carry out a mass shooting or commit an act of terrorism. Whether this conversation meets the legal elements is for prosecutors to prove and a court to decide.
The boundary ordinary users should remember
The case forms a neat contrast with another lawsuit: British Columbia has sued OpenAI, alleging its safety team once flagged a shooter's violent conversations with ChatGPT and did not alert police. In one case the platform reported and the user was arrested; in the other it did not report and was sued afterwards — platforms can land in the spotlight from either direction. The practical conclusion for ordinary users is concrete: do not treat a chatbot as a diary only you can read. What you write into Claude or ChatGPT can be detected by safety systems, read by human reviewers and, when the platform judges an emergency risk to exist, handed to law enforcement; it is not protected by anything like attorney-client privilege. The same company this week also rolled out Claude in-country inference in India, built around keeping data inside one country — where data sits and who may see it are the two questions to settle before handing content to an AI.
