Anthropic Reported a Florida Woman's Claude 'Diary' to Police. She Now Faces a Felony.
A Florida woman is facing felony charges after writing in Claude what she describes as a personal diary and including a threat to attack the Lee County Sheriff’s office. The case is the latest in a series of law enforcement referrals by Anthropic — and the most publicly visible one linked to a diary-style interaction.
What Happened
On September 26, Carli Michelle Heller, of Bonita Springs, Florida, wrote in Claude that she planned to “shoot up” the Sheriff’s office. Heller later told investigators she used the chatbot like a diary.
Claude’s automated safety systems flagged the entry and escalated it to a human reviewer. After determining the threat was credible, the reviewer reported it to law enforcement. Deputies visited Heller’s home, detained her without incident, and transferred the case to an LCSO intelligence detective.
Heller now faces a charge of making a written threat of violence. Under Florida Statute 836.10, transmitting a written or electronic record that threatens to kill or injure someone — in a manner in which another person may view it — is a second-degree felony.
Anthropic’s Disclosure Policy
Anthropic’s terms permit sharing user information in limited emergencies when the company believes disclosure is necessary to prevent death or serious physical injury. Anthropic has exercised this policy before: in August 2026 a San Antonio man was arrested on a felony terroristic threat charge after Anthropic referred Claude queries about an elementary school shooting to authorities. The Florida case follows the same playbook, applied to a threat against a named law enforcement facility.
The incident highlights an assumption many users carry into AI interactions: that conversations are private. They are not. All major providers maintain some form of safety monitoring. Anthropic’s system combines automated flagging with human review — the same pipeline used here.
The Contrast With OpenAI and the FSU Shooting
The case lands in the middle of active litigation about AI provider liability. Florida sued OpenAI in June 2026 over claims that ChatGPT contributed to the April 2025 Florida State University shooting, carried out by Phoenix Ikner. British Columbia filed a separate suit in September 2026 over the February 2026 Tumbler Ridge Secondary School shooting — in that case, OpenAI had flagged and banned shooter Jesse Van Rootselaar’s account in June 2025 but did not alert police because the conversations did not meet the company’s threshold for legal referral.
Across both cases, OpenAI did not make a proactive law enforcement referral. Anthropic appears to have applied a lower threshold. The Heller threat was explicit, targeted a named law enforcement facility, and was escalated after human review — not acted on by automated systems alone.
What It Means
AI systems are not private diaries. Every major provider operates some form of content monitoring, and most explicitly reserve the right to share content with law enforcement when they perceive an imminent safety risk.
Heller’s case adds a specific data point: Anthropic will act on that policy when a human reviewer deems a threat credible, even when the content appears in what a user considers a personal journal. The charge Heller faces carries potential prison time under Florida law.
The industry has not converged on where the referral threshold sits — the FSU case and the Florida arrest illustrate opposite ends of the spectrum. That gap will likely be litigated in both Florida and British Columbia in the months ahead.