A Florida woman is facing a felony charge after writing threats against her local sheriff’s office inside Anthropic’s Claude chatbot — a space she described as her “diary” — and the company’s safety systems escalated the entry to a human reviewer, who reported it to law enforcement. According to the arrest report covered by TechSpot, Carli Michelle Heller, 30, of Bonita Springs wrote on 26 September that she would “shoot up” the Lee County Sheriff’s Office. Deputies identified her through the platform, detained her without incident, and she now faces a charge of making a written threat of violence under Florida Statute 836.10 — a second-degree felony.
🔍 THE BOTTOM LINE: The case is the first high-profile arrest turned on a chatbot company reporting its own user to police, and it punctures the assumption — widely held, widely wrong — that conversations with an AI chatbot are private. Anthropic’s own policy reserves the right to share user information in limited emergencies where it believes disclosure is necessary to prevent death or serious physical injury, and this arrest is that clause firing in anger for the first time.
What the arrest report says
Per the arrest account summarised by regional broadcaster WJCL, investigators say Heller wrote “I’m going to shoot up the sheriff’s right the [expletive] now” on the morning of 26 September, then followed up the next day with “This is 100% last chance I’m done. I got a new [expletive] gun today.” Claude’s automated safety systems flagged the entries and escalated them to a human review team, which assessed the threat as credible and notified law enforcement — as Gadget Review reconstructs from the same report. Deputies responded to Heller’s residence, detained her, and an intelligence detective with the Lee County Sheriff’s Office took over the investigation. A court date was reportedly set for November.
To be plain about the legal posture: the allegations have not been proven in court, and Heller is entitled to the presumption of innocence. The charge itself, though, turns on the words she typed — and those words are not disputed to exist. Florida law criminalises transmitting any written or electronic record threatening death or injury so long as it is made in a manner another person can view it. The twist in this case is who “another person” turned out to be: not a tipster, not a family member, but a contractor at Anthropic reviewing a flagged transcript.
Sheriff Carmine Marceno’s office put the lesson bluntly: “Artificial intelligence is a powerful tool, and like any technology, it can be misused. When someone uses AI to make or facilitate a threat, we have to take that threat seriously. Users need to understand that you are never truly anonymous,” WJCL reports.
The safety escalation chain, and why it exists
Chatbot safety review is not new — OpenAI, Anthropic and Google all run pipelines in which automated classifiers flag content for human contractors. What is new is a completed escalation from classified transcript to human judgment to police report to arrest. Anthropic’s policy language says it may share user information “in limited emergencies” when it believes disclosure is necessary to prevent death or serious physical injury. The company has never hidden that clause; until this week, most users had simply never heard of it.
The context matters, because the industry has been walking this line under pressure. British Columbia is suing OpenAI over the Tumbler Ridge school shooting on the claim that the company could have prevented a mass shooting after its safety team flagged the shooter’s conversations but never alerted police — OpenAI’s position was that the conversations did not meet the threshold for legal referral. Florida itself sued OpenAI in June, alleging ChatGPT contributed to real-world harms including the 2025 Florida State University shooting. Families have sued OpenAI over a school shooting ChatGPT allegedly talked a teen through. Anthropic, in other words, made a different judgment call than OpenAI did in the BC case — and the day after it did, a county in Florida took one person off the street.
Not a private diary — and the timing is the tell
The comparison everyone reaches for — “but it’s my diary” — fails for a structural reason, and TechSpot’s write-up makes the point cleanly: a paper diary has no vendor, no terms of service, no safety pipeline, and no reviewer halfway around the world reading flagged pages. A chatbot conversation is a service interaction with a third party that retains, reviews, and — under stated emergency conditions — forwards what you write. The same week’s adjacent reporting made the point twice over: human contractors reviewing Microsoft Copilot’s image editor can see users’ prompts and uploads, and a Florida man was arrested over a separate AI-generated threat image against the same sheriff.
The timing is also a tell. On 5 October The Verge picked up the same arrest — evidence the escalation chain is now attracting national coverage, not just local reporting — and in the same week Lee County deputies separately arrested a man over an AI-generated image depicting an attack on Sheriff Marceno, per WJCL’s reporting. AI labs are simultaneously under fire for over-refusal and under-refusal, and the legal system is now supplying the answer to which way the incentive points: report, and face a privacy controversy; don’t report, and face a Tumbler Ridge lawsuit.
Our take: for New Zealand readers the practical takeaway is narrower and more useful than the American constitutional debate. Nothing typed into a hosted chatbot should be assumed confidential — that is already the advice given for workplace tools, and this arrest extends it to personal use. The Privacy Act applies to NZ agencies handling personal information, not to a US lab’s emergency disclosure policy, so the protection is effectively whatever Anthropic’s terms say it is. If someone in your life is using a chatbot as an outlet for dark thoughts, the safest advice is no longer just clinical — it is procedural: those words can leave the room.
❓ FAQ
Did Anthropic report the user directly to police? Yes, according to the arrest report as summarised by TechSpot: Claude’s safety systems flagged the entries, escalated them to a human reviewer, who judged the threat credible and notified law enforcement, after which Lee County deputies detained Heller at her home.
What is the user’s legal exposure? A charge of making a written threat of violence under Florida Statute 836.10, a second-degree felony, per the arrest report covered by WJCL. The allegations have not been tested in court; a reported court date was set for November.
Is chatbot conversation private? Not in the way a diary is. Anthropic’s policy permits sharing user information in emergencies where it believes disclosure is necessary to prevent death or serious physical injury — a clause that this arrest shows is actively used, as Gadget Review details.
Has this happened before? Not at an arrest, publicly. The closest precedent is British Columbia’s lawsuit against OpenAI over the Tumbler Ridge school shooting, which alleges OpenAI’s safety team flagged the shooter but did not refer the conversations to police.
🔍 THE BOTTOM LINE
The Anthropic clause about emergencies was never theoretical — it has now produced an arrest. Every chatbot conversation should be treated as potentially readable by the vendor’s reviewers and disclosable to authorities, because the alternative assumption got a 30-year-old Florida woman charged this week. The safety-versus-privacy trade AI companies have been quietly pricing for two years is being settled one escalation at a time, and users are the last to be told which side their provider chose.
📰 Sources
- TechSpot — Florida woman used Claude as a diary, then Anthropic reported an entry to police
- Gadget Review — Florida Woman Used Claude as a Diary. Anthropic Reported Her to Police.
- WJCL / Gulf Coast News — Florida woman accused of telling Claude AI she was going to ‘shoot up the sheriff’s office’
- The Verge — Florida woman arrested for allegedly making threats in an AI chat
