In late September, Carli Heller used an AI chatbot to vent anger at the Lee County Sheriff's Office, writing that she planned to shoot up the office and later mentioning a new gun. Anthropic's safety systems flagged the messages and escalated them to the FBI, which alerted the sheriff's office. Heller was arrested and charged with making a written or electronic threat of a mass shooting, a felony under Florida law; she has pleaded not guilty.
The central legal question is whether a statement typed into a chatbot can count as a criminal threat. Under Florida law, as interpreted in T.R.W. v. State, prosecutors must show the defendant transmitted the communication for the purpose of issuing a threat or with knowledge that it would be viewed as one. The U.S. Supreme Court's later decision in Counterman v. Colorado allows conviction if the defendant was reckless about that possibility, meaning they consciously disregarded a substantial risk that a human would see and perceive the message as threatening.
Writing in Reason, Eugene Volokh argues that Heller likely assumed she was communicating only with software, not with a human reviewer. If that is true, she lacked the required mental state, and the charge should fail. He also notes that even under the recklessness standard, the state would need to prove beyond a reasonable doubt that Heller was aware her words might reach and alarm a person—something he considers unlikely based on the reported facts.