Florida v. Zhou (ChatGPT Murder-Threat Rehearsal and OpenAI FBI Referral)

Darren Zhou, a 25-year-old Goldman Sachs analyst living in South Palm Beach, Florida, turned to ChatGPT in March 2026 after his girlfriend of six months ended their relationship. Court records show the conversations began as breakup processing — her hobbies, the places she frequented, his jealousy of other men, his desire to reconcile — and quickly escalated into detailed plans of rape, murder, and murder-suicide.

According to investigators, Zhou wrote 'I'm gonna kill her by the end of this month' and 'If I can't have her then nobody can.' He described waiting for his ex-girlfriend at a gym with flowers and killing her if she rejected him, and forcing her at gunpoint to his home to sexually assault her and hold her hostage. He told the chatbot he had access to an AR-15-style rifle, a Glock handgun, and a 12-gauge shotgun (his attorney later said he owned no firearms), and claimed to have sent the woman photographs of guns, zip ties, and latex gloves. His stated targets eventually grew to include her family.

OpenAI's monitoring detected the threats and the company reported Zhou to the FBI — its policy allows a small human review team to examine conversations flagged for planned harm to others and, in serious cases, to tip law enforcement. Federal agents turned over roughly two months of chat logs to the Palm Beach County Sheriff's Office in May 2026. A deputy who reviewed them said they were not 'vague emotional outbursts' but a pattern of rehearsal and planning. The released logs contained only Zhou's side of the conversations, not ChatGPT's responses.

Deputies performed a welfare check on the 21-year-old woman, who said she had ended the relationship over Zhou's 'erratic, jealous and controlling behavior' — by phone, because she feared his reaction — and that he had since bombarded her with calls, texts, and social media messages from app-generated numbers, many harassing, sexual, or threatening. Zhou was arrested in May 2026, charged in June, and pleaded guilty on August 13, 2026 to aggravated stalking, written threats to kill, and unlawful use of a two-way communications device. The court withheld adjudication and imposed eight years of probation; the judge said he accepted the deal only because the victim approved it.

The case is believed to be among the first publicly documented instances of an AI company's own threat-detection pipeline producing a law-enforcement referral, arrest, and conviction. The harm tracked here is the stalking and threat campaign against the ex-girlfriend; the intended physical harm was interdicted.

• No harmful AI output is documented in this case: the chat logs released to investigators contained only Zhou's messages, and no source alleges ChatGPT encouraged, validated, or assisted the plans
• The chatbot served as the private venue where a rejected ex-partner rehearsed rape and murder plans over roughly two months, in escalating detail
• OpenAI's abuse-monitoring systems flagged the conversations, routed them to human review, and generated an FBI tip — the provider-side detection pathway functioned as designed
• The 'written threats to kill' conviction rests on messages Zhou wrote to the chatbot itself, treating statements made inside an AI conversation as criminally cognizable threats

The harm to the victim was a months-long stalking and threat campaign by a rejected ex-partner, which ran in parallel with — and was documented inside — his ChatGPT conversations.

The AI relationship functioned as a private rehearsal space: Zhou iterated on specific, operational plans (locations, weapons, sequences) conversationally, in a venue he apparently believed was consequence-free. The escalation from grievance to operational planning is preserved verbatim in the logs.

The same relationship became the interdiction point. OpenAI's detection and FBI referral triggered the welfare check and arrest that ended the campaign before physical harm occurred. This incident is entered with that dual character stated plainly: the participant relation carried the threat conduct, and the provider's monitoring of that relation is what stopped it.

• March 2026: Zhou begins using ChatGPT after the breakup; conversations escalate from breakup processing to rape, murder, and murder-suicide planning
• May 2026: OpenAI reports the threats to the FBI; federal agents turn over roughly two months of chat logs to the Palm Beach County Sheriff's Office; deputies perform a welfare check on the ex-girlfriend and arrest Zhou, who posts $100,000 bail after two days in jail
• June 2026: Assistant State Attorney Ana Cuskova charges Zhou with aggravated stalking, written threats to kill, and unlawful use of a two-way communications device (felonies carrying up to 25 years)
• August 13, 2026: Zhou pleads guilty to all three charges under a negotiated deal approved by the victim; Circuit Judge Scott Suskauer withholds adjudication and sentences him to eight years of probation, including no contact with the victim and a batterer intervention program
• Zhou was fired by Goldman Sachs; his attorney attributed the messages to 'a very difficult mental health episode'