ITHome 10/5 news: A 30-year-old woman in Lee County, Florida, USA was arrested on suspicion of writing violent threats against the local sheriff's office in a conversation with Anthropic's chatbot Claude.
According to local police, Carli Michelle Heller, a resident of Bonita Springs, said in a conversation on 9/26 that she would shoot up the Lee County Sheriff's Office. The content immediately triggered Anthropic's safety risk control mechanism and was reported by Claude to human review and law enforcement.
According to local media WINK News, Sheriff Carmine Marceno told the outlet that Heller later said she used the AI as a "diary."
Investigators said Anthropic monitors keywords and potential threat content through safety mechanisms; relevant information is escalated to human review depending on severity, and after confirmation by the human team, local law enforcement is contacted.
After receiving the report, local officers went to her residence and took the woman into custody without incident; the case was then handed over to an intelligence detective from the Lee County Sheriff's Office. Heller told investigators that she currently faces one charge of written threats to commit violence, but has not yet been convicted by a court.
Florida Statute 836.10 provides that any person who threatens to kill, inflict bodily harm, conduct a mass shooting, or commit an act of terrorism through a written or electronic record that can be viewed by others may be guilty of a second-degree felony. The current statute also makes clear that the provisions apply to electronic records, and states that sending, posting, or transmitting a record containing such threats to others may constitute a crime. Whether the Claude conversation in this case meets the specific legal elements, and whether prosecutors can prove the facts of the crime in judicial proceedings, remains to be decided by the court.
Marceno said AI may have changed the way threats are made, but the commitment to protecting Lee County residents remains unchanged, and when someone uses AI to carry out or assist in carrying out threats, police must take it seriously.
Meanwhile, the case has also sparked discussion about the privacy boundaries of chatbots. Anthropic's currently published government information request policy shows that the company generally only discloses user information to the government after receiving valid legal process, but there is an emergency disclosure exception if the company believes an emergency situation could lead to imminent bodily harm or death, and timely provision of relevant information could avert the danger.
Anthropic's transparency page, updated in 7/2026, also shows that the company uses a safety team responsible for designing and implementing detection and monitoring mechanisms to enforce usage policies, and for handling data requests from law enforcement agencies and governments, including emergency disclosures. From this, using Claude as a private diary does not mean the content will necessarily be seen only by the user; the platform may detect and review some content in accordance with safety and usage policies, and under the corresponding conditions may also provide information to law enforcement agencies.
In addition, Anthropic's terms of service also mention that conversations with the AI are not protected by a privilege similar to attorney-client confidentiality, but Anthropic has not yet detailed the review of this conversation.
Beyond this case, the Canadian province of British Columbia recently sued OpenAI and its CEO Sam Altman in a US federal court in San Francisco, alleging that OpenAI had previously flagged the February (2) Tumbler Ridge school shooter's violence-related conversations with ChatGPT but did not report them to police, and that the incident could have been prevented.
According to TechSpot, Florida already sued OpenAI and Altman in June (6), alleging that ChatGPT caused real-world harm, including the 2025 Florida State University shooting (ITHome note: the above are the plaintiffs' allegations; the courts have not yet determined any liability).