According to the arrest report cited by multiple outlets, Heller wrote in a Claude chat on September 26, 2026, that she planned to "shoot up" the Lee County Sheriff's Office. The next day, she wrote that she had acquired a new gun and that this was her "last chance," language investigators treated as a credible threat.
From automated flag to police referral
Anthropic's automated monitoring systems flagged the conversation as potentially dangerous and escalated it to the company's human review team. After reviewers assessed the messages as a credible threat, Anthropic used its emergency-disclosure process to contact the Lee County Sheriff's Office and shared information identifying the user.
Deputies detained Heller at her Bonita Springs home without incident. She has been charged under Florida Statute 836.10 with making a written threat of violence, a second-degree felony under Florida law that covers electronic or written threats to kill or injure. A court hearing has been scheduled for November, and the facts outlined in the arrest report have yet to be proven in court.
"Never truly anonymous"
Heller told investigators she had been using Claude as a personal diary, Lee County Sheriff Carmine Marceno said. But the diary analogy ended where the threat began: a message about attacking a specific, recognizable target is not private venting once a safety system picks it up.
"Users need to understand you are never truly anonymous, specifically AI chatrooms and searches," Marceno said, warning that AI may be changing how threats are created, but law enforcement treats them the same as any other.
Anthropic's privacy policy allows the company to share user data with law enforcement when it reasonably believes disclosure is necessary to prevent serious harm or death. Conversations flagged by its security systems may also be reviewed by company staff, even when a user has opted out of having their data used for model training. Reporting threats, child sexual abuse material, and self-harm intent to law enforcement is standard practice across technology companies, not unique to Anthropic.
The third chat-to-police referral since August
The case is at least the third reported AI-chat-to-police referral since August, underscoring a pattern of AI companies escalating certain flagged conversations to authorities — and intensifying scrutiny of how AI firms detect threats, how human reviewers access private conversations, and what liability they carry for acting, or failing to act.
OpenAI made a similar promise after a school shooting in Canada, saying it would notify police about ChatGPT conversations indicating an imminent threat. In Florida, a separate criminal investigation has examined an attacker's use of ChatGPT before a university shooting. Meanwhile, lawmakers including Senators Josh Hawley and Chris Murphy are pushing legislation that could expose AI companies to lawsuits over their platforms' role in such incidents.
For the millions of people who treat chatbots as confidants, the Bonita Springs case draws a sharp line: the system does not read a user's intent. It reads words and context — and then a person decides whether those words are a concrete threat. When that judgment goes against the writer, a private diary entry can become police evidence.
By Hamza Faisal
