Threats Typed Into Claude Lead to Charges After Anthropic Alerts Police

In Florida, threatening remarks typed into Claude led to a police report and criminal charges. According to an arrest report cited by WINK News, Anthropic alerted police after the conversation went through the company’s safety screening process and a human review. The woman was arrested and charged with making written threats. She has not been convicted.

What was in the conversation

Investigators allege that on September 26 the woman typed a message into Claude saying, in effect, that she would attack the Lee County Sheriff’s Office. The next day, they say, she wrote that she had obtained a new gun. The sheriff said that any use of AI that appears to involve a threat has to be taken seriously.

The sheriff told local media that the woman had been using Claude like a diary. The question now is whether the exchanges that investigators read as threats meet the legal requirements for criminal liability.

Florida Statute 836.10 makes it a second-degree felony to send an electronic record containing a specific threat in a manner that allows another person to see it. The statute covers threats to kill or injure, as well as threats of a mass shooting or an act of terrorism. Whether a conversation held with Claude falls under that provision will be settled in court.

Who can see my conversations

Anthropic says that designated members of its safety teams may review conversations when doing so is necessary to enforce its Usage Policy, the rules that govern how its services may be used. In other words, what a user types into the AI can become subject to human review under those conditions.

The company’s privacy policy also spells out when it may hand personal information to law enforcement. Anthropic states that it may share information if it believes in good faith that doing so is reasonably necessary to prevent serious harm to people or property.

Why this matters for users in Korea

Korea is among the regions where Claude.ai and Anthropic’s developer services are officially supported. That gives Korean users reason to look closely at how conversations are reviewed and under what conditions information is passed on, especially when they are typing private thoughts into a chatbot. OpenAI, which operates ChatGPT, has likewise said that it refers cases to law enforcement when it judges a risk of harm to others to be imminent and credible.

Two things are worth watching from here. One is how the court applies the written-threats statute to a conversation held with Claude. The other is whether Anthropic explains the reasoning behind this particular report. What the case makes clear is that a private conversation with an AI can pass through safety review and end up as a criminal investigation.