Florida Woman Arrested After Anthropic Flags Mass Shooting Threat in Claude AI Chat
Carli Michelle Heller was arrested in Lee County, Florida, after Anthropic flagged her Claude chat session containing threats to shoot up a local sheriff's office, verified the threat via human review, and notified police. Heller told investigators she had been using the AI chatbot as a personal journal. This case highlights the tension between user privacy expectations and AI safety guardrails, demonstrating that conversations with LLMs are monitored and subject to emergency law enforcement disclosures. It also contrasts with recent lawsuits alleging that AI companies failed to report flagged threats prior to real-world violent events. Anthropic's terms state that AI chats are not protected by legal confidentiality privileges and permit emergency disclosures without prior legal process if imminent physical harm or death is suspected. Heller faces charges under Florida Statute 836.10 for making written threats of violence, a second-degree felony that applies to electronic communications.
## BACKGROUND
Major Large Language Model (LLM) providers employ automated keyword monitoring alongside human review teams to enforce usage policies against violent or harmful content. While AI platforms typically require valid legal process like subpoenas to release user data to governments, most maintain emergency disclosure exceptions for imminent threats to human safety.