Anthropic handed a user's diary entry over to the police

⚡ The gist in 5 seconds - Anthropic reported a Claude chatbot entry to the police: a user wrote that she planned to "shoot up" a sheriff's office. - The entry was flagged by the safety system and reviewed by a human reviewer, who decided it was a real threat. - The woman faces a charge of written threat of violence — under Florida law, this is a second-degree felony. ### 🔍 What was found On September 26, Carly Michelle Heller of Bonita Springs, Florida, used Claude as a personal diary and wrote that she would attack a sheriff's office. She later said she kept a diary in the Anthropic chatbot. Claude's safety systems sent the entry for human review. After assessing it, the reviewer passed it on to law enforcement. Police identified Heller, visited her home, and detained her without incident. She was charged under Florida Statute 836.10, which classifies as a second-degree felony sending or transmitting a written or electronic record threatening to kill, commit a mass shooting, or commit terrorism. The record must also be written in a way that another person could see it. Anthropic states that in limited emergency situations it may disclose user information when necessary to prevent death or serious physical injury. ### 💡 Why it matters This case is yet another reminder that chatbots are not a private diary. Even if users consider their entries personal, they can be viewed by company employees and handed over to authorities. Anthropic clearly does not take risks when it comes to potential threats. For users, this means that any discussions of violence or unlawful actions in AI services can have legal consequences. The company's decision to hand the entry over to the police shows how AI companies interpret their responsibility for preventing harm. ### 🧩 Context Similar situations have already drawn attention to the role of AI companies in preventing violence. Last month it was reported that OpenAI and Sam Altman had become defendants in a lawsuit from British Columbia: the company is accused of being able to prevent a mass shooting. The shooter was flagged by OpenAI's safety system for conversations about gun violence, but the company did not notify the police because the conversations did not reach the threshold for reporting. In June, Florida also filed a lawsuit against OpenAI and Altman, claiming that ChatGPT contributed to real harm, including the 2025 shooting at Florida State University. Separately, it was reported that human reviewers of Microsoft Copilot can see users' prompts, uploaded photos, and AI edits — some tasks contain potentially unlawful material.

🔗 Read on www.techspot.com