Think twice before treating an AI chatbot like a private diary. A Florida woman allegedly used Claude like a diary while writing threats against a sheriff's department, with an arrest report describing how the messages were flagged by the platform's safety systems before staff reviewed them and authorities became involved.
The case does not establish that Claude routinely reports private conversations to police. But it highlights how messages entered into an AI chatbot can be subject to automated safety checks and human review, with Anthropic saying user information may be disclosed to law enforcement in limited emergency circumstances.
A Claude Chat Triggered A Police Alert
Investigators obtained an arrest report detailing how 30-year-old Carli Michelle Heller allegedly set off the sequence by typing on 26 September that she intended to 'shoot up' the sheriff's office.
According to the report, another message appeared early the following morning in which the user mentioned getting a new gun and described the situation as a 'last chance'.
According to the arrest report, the messages were flagged by Anthropic's AI safety and security systems and, because of the severity of the statements, escalated to a human review team, which then reported them to law enforcement.
Sheriff Carmine Marceno later told WINK News that Heller said she used AI like a 'diary'.
Anthropic has not publicly confirmed the precise handling of Heller's specific conversation, so the reported sequence should be attributed to the arrest report rather than presented as a case-specific account from the company.
AI Safety Systems Flagged The Messages
Anthropic's public documentation describes a layered safety framework combining automated detection systems with human review to identify potentially dangerous activity on its platform.
The company says its safeguards use specialised classifiers to identify specific types of policy violations in real time, alongside human review and other enforcement measures.
Anthropic says these systems can be used to detect harmful activity and determine whether enforcement action is necessary. Its public documentation therefore broadly resembles the process described by investigators in Heller's case, although the company has not independently detailed which specific safeguards or review procedures were used for her conversation.
Anthropic Can Share Data In Emergencies
Anthropic's public policy says it generally provides user information to government authorities only when required by valid legal process, but makes an exception when it believes there is an emergency involving imminent physical harm or death and providing information without delay may help avert it.
The company has also said that, in limited circumstances, user data may be disclosed to law enforcement in response to emergency requests involving a danger of death or serious physical injury.
That policy provides broader context for how a potentially dangerous AI conversation could lead to law-enforcement involvement in exceptional circumstances. It does not, by itself, establish that this specific case was handled under Anthropic's emergency-disclosure procedure.
Police Arrived At Her Home
Following the alert described in the arrest report, deputies went to Heller's residence on Dean Street in Bonita Springs, where she was detained without incident, according to local reporting.
Authorities charged Heller with making a written threat of violence under Florida law.
Under Florida Statute 836.10, it is unlawful to send, post or transmit a writing or electronic record in a manner in which it may be viewed by another person when it contains a threat to kill or cause bodily harm, or a threat to conduct a mass shooting or act of terrorism. A violation is a second-degree felony.
The allegations have not been proven in court. Heller is due to appear in court in November, according to WINK News.
AI Chats May Not Be Private Diaries
The case highlights the tension between the conversational way people interact with AI and the safety systems operating behind those conversations.
FITECH CEO Ian Marlow told WINK News that the conversational nature of AI can make users feel an interaction is more private than it actually is.
Dr Chrissann Ruehle, an assistant professor of AI and management at Florida Gulf Coast University, said people may turn to AI because a machine can feel non-judgemental, but warned against treating the technology as a private diary.
'AI is not your best friend,' Ruehle said. 'We need to keep in mind this is a technology.'
Claude is not a locked diary, and Anthropic's published safeguards make clear that the company uses automated systems and human review to enforce its policies.
That does not mean every personal or sensitive conversation is reported to police. But this case shows why users should understand that an AI service operates under its own safety and privacy rules rather than functioning like a private journal.
