Carli Michelle Heller of Bonita Springs, Florida, was arrested on a felony charge after Anthropic's human review team reported her interactions with the Claude AI chatbot to police. According to the arrest report, Heller, 30, wrote on September 26 that she would "shoot up" the Lee County Sheriff's Office (LCSO). The following day, a message in the chat allegedly mentioned a new gun. Heller informed investigators that she used Claude as a diary.
Anthropic monitors chatbot conversations for phrases and content that could be deemed threatening. Depending on the severity of the content, these conversations can be escalated for human review. In Heller's case, the human review team determined the content warranted reporting to law enforcement. Lee County Sheriff Carmine Marceno stated that users are "never truly anonymous" when using AI chatbots.
Deputies identified Heller using information provided by Anthropic and detained her at her home without incident. An LCSO intelligence detective subsequently took over the investigation. Heller faces a felony charge under Florida Statute 836.10, which addresses written or electronic threats of a mass shooting or act of terrorism. This statute makes transmitting an electronic record threatening to kill or injure another person, conduct a mass shooting, or commit an act of terrorism a second-degree felony when the record can be viewed by another person.
Anthropic's public privacy policy, effective September 10, states that disclosure to law enforcement may occur when the company has a good-faith belief that disclosure is reasonably necessary to "prevent serious harm to any person or to property." The company's transparency policy indicates that it processes data requests from law enforcement agencies "in accordance with applicable laws while protecting user privacy." The company's emergency-disclosure policy allows for bypassing legal processes when imminent physical harm is believed to be likely.
This incident marks at least the third such conversation involving an Anthropic chatbot to reach police since August. In one prior incident in San Antonio, an August 11 chat with an Anthropic AI involved queries about shooting at an elementary school, leading to an arrest on a felony terroristic threat charge after the FBI alerted police. Days later, on August 14, a user in San Francisco threatened Anthropic CEO Dario Amodei in a Claude chat, claiming to have purchased an AR-15. Anthropic notified police in that instance, but the user was not arrested or charged. OpenAI has also reported user threats to law enforcement, including a case in March where a ChatGPT user discussed killing his ex-girlfriend, resulting in an FBI report and a guilty plea for written threats to kill.
Anthropic's privacy policy, updated to be effective September 28, 2025, allows for the use of user-generated data, including chats, to train and improve its AI models, with an opt-out option for users. If users permit data usage, it is retained for five years for training purposes. The company has stated it tracks "concerning behavior over time" through a "person-of-interest process" to "catch escalation patterns early."