OpenAI Reported ChatGPT Murder Plot Conversations to FBI

OpenAI's automated monitoring system detected a detailed murder plot in ChatGPT conversations and referred the case to the FBI before any crime took place. The case of the 25-year-old former Goldman Sachs analyst from Florida reached court in August, where the man pleaded guilty.

What OpenAI's system detected

Zhou began conversing with ChatGPT in March 2026, shortly after breaking up with his 21-year-old girlfriend from Lake Worth, Florida. Over more than two months, he built up a detailed profile of the woman in the chat, describing her interests, daily schedule, and the places she regularly visited, including her gym.

OpenAI's monitoring system automatically flagged the conversations as potentially dangerous and routed them for human review. Under the company's policy, in cases indicating planning of serious harm, staff can refer the matter to law enforcement. That is what happened here: in May 2026, OpenAI notified the FBI and handed over the full chat logs.

The content of the threats

I will kill her by the end of this month - Darren Zhou, in a conversation with ChatGPT

If I can't have her, no one will - Darren Zhou, in a conversation with ChatGPT

According to court documents, Zhou's messages went beyond emotional venting and described a concrete plan of action, including waiting with flowers in the parking lot of the gym his ex-girlfriend attended, and how he would react if rejected. The victim herself independently contacted the FBI, saying she had broken up with Zhou over the phone out of fear of his reaction.

Charges and sentence

The material from OpenAI reached the Palm Beach County Sheriff's Office, and the investigation led to Zhou's arrest. After being arrested, he was released on $100,000 bail but lost his job at the investment bank. In June 2026, he was formally charged with aggravated stalking, written threats to kill, and unlawful use of a communication device.

On August 13, 2026, Zhou pleaded guilty in court. Judge Scott Suskauer withheld adjudication of guilt and imposed eight years of probation, two of which the man must serve with an electronic monitoring bracelet. The conditions also include a no-contact order with the victim, a ban on possessing weapons, and participation in a batterers' intervention program.

What this means for AI safety policy

The Zhou case is one of the first widely reported instances in which a major AI provider's automated system directly led to the prevention of a planned violent crime before it happened. OpenAI has spent months expanding its mechanisms for detecting conversations that indicate a risk of self-harm or harm to others, in response to earlier criticism over chatbots' role in users' mental health crises.

For companies developing large language models, the case demonstrates the practical operation of a policy on reporting content indicating planned serious harm to law enforcement, something OpenAI and other providers have long stated they do but have rarely documented with a concrete, adjudicated case. It is also a reminder that conversations with AI chatbots are not entirely private when they contain credible signs of a threat to life.

The case was heard in the United States and has no direct bearing on Polish law, but it illustrates the direction global AI providers, including those available on the Polish market, may be heading. The question of exactly how conversation-monitoring systems should balance user privacy against violence prevention remains open and will keep resurfacing in similar cases.