Carli Michelle Heller of Bonita Springs faces a felony charge after a human review team at Anthropic read her Claude conversations and contacted law enforcement. Investigators state that Heller wrote on September 26 that she would shoot up the Lee County Sheriff's Office. Additional platform records showed that she claimed she had acquired a new gun by the next day.
Anthropic operates automated platform monitoring designed to flag key phrases and threatening content. Due to the severity of the statements in this case, the chats went to a human review team. That team reported the findings to law enforcement, leading deputies to identify Heller, visit her home, and detain her without incident. She now faces a formal charge of making a written threat of violence.
Terms reserve the right to report chats
Anthropic consumer terms outline the company's authority to handle user inputs, outputs, and actions. The platform reserves the right at its sole discretion to report user data to law enforcement agencies. A privacy policy effective September 10, 2026, permits data sharing with police whenever the company maintains a good-faith belief that disclosure is reasonably necessary to prevent serious harm.
Users who opt out of data handling for model training still face retention periods of three months rather than five years. Conversations flagged for safety reviews also continue to serve as training material to improve the company's ability to detect harmful content.
Other firms handle threats similarly
Heller is not the first person to face police referrals over chatbot prompts. The San Francisco Standard reported in September that Anthropic referred another user to police after he allegedly wrote that he bought an AR-15 and targeted CEO Dario Amodei. An employee told officers he felt no immediate safety concern and declined to show the messages based on company policy. The user later stated he was joking and faced no arrest at the time of the report. An Anthropic spokesperson described the outcome as the safeguards process working as intended while declining to comment on threats against employees.
Other major artificial intelligence providers follow similar reporting procedures. Following a school shooting in Tumbler Ridge, British Columbia, in February that killed eight people, OpenAI stated it weighed alerting Canadian police in June 2025. OpenAI concluded that account did not present an imminent and credible risk of serious physical harm. An August 2025 policy post from OpenAI outlines routing threats against others to human reviewers who may refer them to law enforcement.
