AIBRIEF
Only the user's prompts to ChatGPT, not the chatbot's replies, were shared with Palm Beach deputies, and a deputy there called two months of messages 'rehearsal, not venting.'
A Florida man was arrested after telling ChatGPT he planned to kill his ex-girlfriend, court records show. A small team inside OpenAI reviewed what he had typed, alerted the FBI, and Palm Beach County Sheriff's Office deputies later called two months of messages "rehearsal, not venting."
The case is among the first public examples of OpenAI's published threat-review policy firing on its own initiative. Under that policy, a small internal team may review flagged ChatGPT interactions when a user appears to plan violence against others, and in serious cases may tip law enforcement. The handoff ran from OpenAI to the FBI, which routed the material to Palm Beach County Sheriff's Office deputies; they performed a welfare check on the ex-girlfriend and then arrested the suspect.
Unlike a routine subpoena response, the case turned on two details. Only the user's prompts reached police, not the chatbot's replies. A PBSO deputy characterized the logs as "a pattern of rehearsal and planning" rather than "vague emotional outbursts." Court records preserve direct threat quotes: "I'm gonna kill her by the end of this month," and "If I can't have her then nobody can."
The 21-year-old ex-girlfriend, from Lake Worth Beach, told deputies she ended the roughly six-month relationship over what she described as his "erratic, jealous and controlling behavior," and that he later harassed her through app-generated numbers, texts, DMs, and TikTok comments. Targets later expanded to include her family. Final federal indictment counts and the assigned assistant U.S. attorney are not visible in the available source excerpt.