Arrest for words: ChatGPT read threats and called the police

OpenAI handed over to the FBI a user's chat with ChatGPT in which he described his intention to kill his eight-year-old son and discussed attacks on schools and churches. The man was arrested a day before the planned murder. However, OpenAI refused to provide investigators with ChatGPT's responses, raising questions about the prospects for prosecution.

OpenAI handed over to the FBI a user's chat with ChatGPT in which he detailed his intention to kill his eight-year-old son and discussed attacks on schools and churches. The company detected trigger words in the messages and notified the FBI. Bureau agents arrested the man a day before the date he himself named as the day of the murder. During a search of his home, containers with an unknown substance were found. This is the first publicly known case of an arrest for correspondence with AI without the crime itself having occurred. However, OpenAI only provided the user's messages to investigators, refusing to hand over ChatGPT's responses. Lawyers point out that without the full dialogue, it is impossible to establish the context — whether this was a real intent or a therapeutic conversation. The man has not yet been charged. The case has sparked a debate about the boundaries of confidentiality in correspondence with AI.