A Florida man’s conversations with ChatGPT about his ex-girlfriend didn’t stay between him and the chatbot.
OpenAI alerted the FBI after 25-year-old Darren Shida Zhou used ChatGPT to discuss plans to rape and kill his former girlfriend, according to court records reviewed by USA Today. Federal agents eventually provided roughly two months of Zhou’s messages to the Palm Beach County Sheriff’s Office.
“I’m gonna kill her by the end of this month,” Zhou wrote in one message preserved in court records. “If I can’t have her, then nobody can.”
The conversations helped lead investigators to Zhou, who was arrested in May and charged with three felonies. He pleaded guilty Aug. 13 and received eight years of probation under an agreement that kept him out of prison and allowed the judge to withhold adjudication.
OpenAI Alerts FBI About Florida Man’s ChatGPT Conversations
Zhou was working as a financial analyst for Goldman Sachs in West Palm Beach when his six-month relationship with a 21-year-old woman ended.
According to investigators, he began talking to ChatGPT about the breakup in March. He discussed his ex-girlfriend’s hobbies, places she frequented, other men around her, and his desire to get back together with her.
The conversations soon turned to threats. Zhou discussed waiting for his ex-girlfriend in a gym parking lot with flowers and killing her and himself if she rejected him. Other messages described kidnapping and sexually assaulting her, holding her hostage, and killing members of her family, according to court records cited by WPBF and USA Today.
Zhou also wrote about owning firearms and sent his ex-girlfriend photographs of guns, zip ties, and latex gloves. His attorney, Steven Bell, later told the court that Zhou did not actually own firearms.
OpenAI explains on its safety page that conversations involving plans to harm other people can be routed to specialized systems and reviewed by a small team trained on the company’s usage policies.
If those reviewers determine there is an imminent threat of serious physical harm to another person, OpenAI says the company may refer the case to law enforcement.
That’s what happened in Zhou’s case, with OpenAI alerting the FBI.
Deputies Review Two Months of ChatGPT Messages
By May, federal agents had provided the Palm Beach County Sheriff’s Office with about two months of Zhou’s ChatGPT logs.
The material given to a deputy reportedly contained Zhou’s messages to ChatGPT, but not the chatbot’s responses. After reviewing the logs, the deputy concluded Zhou’s statements weren’t simply vague emotional outbursts and instead showed repeated planning and rehearsal, according to USA Today.
Deputies conducted a welfare check on Zhou’s former girlfriend. She told investigators she ended the relationship because of what she described as his “erratic, jealous, and controlling behavior” and broke up with him over the phone because she feared how he would react.
Zhou continued to call, text, and contact her on social media despite repeated requests to stop. Investigators said he also used app-generated phone numbers to reach her after she blocked him.
She saved screenshots of messages investigators described as harassing, insulting, sexual, and threatening. Investigators were then able to compare communications she received with the ChatGPT records provided by federal agents.
Florida Man Is Booked on Three Felony Charges
Zhou was arrested and booked into the Palm Beach County jail on May 29.
The booking record lists charges of written or electronic threats to kill or cause bodily harm, aggravated stalking involving a credible threat, and using a two-way communication device to facilitate a felony.
Those same charges appear in Zhou’s official FINRA record, which identifies the proceeding as Palm Beach County 15th Judicial Circuit case No. 50-2026-CF-004017-AXXX-MB.
USA Today reports Zhou spent two days in jail before posting $100,000 bail.
FINRA also independently confirms Zhou’s employment history. The regulator’s records show he was registered with Goldman Sachs & Co. LLC in West Palm Beach from March 2024 until June 2026 and identify his job as a financial analyst in Asset & Wealth Management.
Goldman Sachs later fired Zhou after learning about the case.
Florida Man Pleads Guilty, Avoids Prison
Assistant State Attorney Ana Cuskova charged Zhou with all three felonies in June. The charges carried a potential combined sentence of up to 25 years in prison.
Zhou instead reached a negotiated agreement and pleaded guilty to all three charges on Aug. 13.
Circuit Judge Scott Suskauer withheld adjudication and sentenced him to eight years of probation. With adjudication withheld, Zhou pleaded guilty but was not adjudicated a convicted felon in the case.
His probation includes electronic monitoring for the first two years, no contact with his former girlfriend, and a prohibition on possessing firearms. Other reported conditions include completing a batterer’s intervention program, undergoing a mental health evaluation and treatment, and staying away from drugs and alcohol.
Suskauer said he accepted the agreement because Zhou’s former girlfriend approved the resolution.
The judge also had some direct advice for Zhou.
“Get over it and move on,” Suskauer told him, according to reports of the sentencing. “There are other people in the world.”
Zhou’s attorney pointed to his client’s lack of a prior criminal record and academic history. Bell said Zhou graduated magna cum laude from Northeastern University.
“He’s not a violent person. He is not a bad person,” Bell said, according to USA Today. “He went through just a very difficult mental health episode and he’s doing unbelievable now.”
OpenAI Says Threats Against Others Can Be Referred to Police
Zhou’s case followed a specific path from his ChatGPT conversations to local investigators: OpenAI reviewed the threatening activity and alerted the FBI, which later provided the records to the Palm Beach County Sheriff’s Office.
OpenAI publicly describes that process on its website. The company says users believed to be planning harm against other people can have their conversations routed for human review. Reviewers are authorized to take action against accounts and may refer cases involving an imminent threat of serious physical harm to law enforcement.
The company draws a line between threats against other people and self-harm cases. OpenAI says it does not currently refer self-harm cases to law enforcement because of privacy concerns surrounding those conversations.
For Zhou, the referral put months of messages in front of federal agents and, eventually, Palm Beach County investigators.
Zhou pleaded guilty to all three charges. Because Suskauer withheld adjudication, he was not adjudicated guilty of the three felonies.