사진 확대

An American man who repeatedly entered plans to rape, kidnap, and kill his ex-girlfriend in the artificial intelligence (AI) chatbot ChatGPT was arrested by OpenAI.

Darren Zhou, 25, who lived in South Palm Beach, Florida, and worked as a financial analyst at Goldman Sachs' West Palm Beach branch, was eventually caught after telling details about his plan to kill his girlfriend in a chat with ChatGPT in March, according to the IT media Puccurism and local newspaper Palm Beach Post on the 14th (local time).

According to the investigation records, Zhou broke up with his 21-year-old girlfriend for six months and began to express his feelings and jealousy after the breakup on ChatGPT, as well as his partner's hobbies and frequent places. The conversation soon turned into a concrete plan of violence.

"I will kill her by the end of the month," he wrote in a message left in court records, "if I can't have it, no one can have it." He explained the scenario in which he would take out a gun and kill himself if he was rejected while waiting with flowers in the parking lot of a sports facility where his ex-girlfriend attends. There were even plans to threaten the opponent with a gun, drag him to his house, rape him, and take him hostage, and say that if the police are dispatched, he will kill himself after shooting several times, and even mention that he will kill the entire family of the opponent.

Zhou told ChatGPT he had purchased an AR-15 family rifle, a Glock pistol and a shotgun, and said he had sent his ex-girlfriend photos of guns, cable ties (zipperties) and latex gloves. There was also a line that said, "I sent you a text saying I would rape you tonight."

사진 확대

In fact, he used various phone numbers and applications to contact his ex-girlfriend persistently and send threat and sexual harassment messages. The victim stated that Zhou's "impulsive, jealous, and controlling behavior" notified her of the breakup by phone, and that she was unable to report it because she was afraid of his reaction and collected the messages in screenshots.

OpenAI deemed the conversation an emergency and reported it to the Federal Bureau of Investigation (FBI). The FBI handed over two months' worth of conversation records to the Palm Beach County Sheriff's Office in May. A sheriff's report that reviewed the records evaluated it as a pattern of repeated planning and rehearsal, rather than a "blind emotional outpouring."

Zhou was arrested in May and detained for two days before being released on bail of $100,000 (about 140 million won). Goldman Sachs confirmed to Futurism that it fired him after learning of the incident in early June.

He was charged with aggravated stalking, writing death threats and illegal use of telecommunications devices for the crime, but instead of pleading guilty in August, he agreed to a Pliwagen (guilty negotiation) that avoids jail time and felony convictions. The court accepted the agreement and suspended the sentence and sentenced it to eight years of probation, citing the fact that the victim had approved it. Zhou wears electronic anklets for the first two years, completes a domestic violence perpetrator intervention program, undergoes mental health evaluation and treatment, and is prohibited from contacting his ex-girlfriend, guns, drugs, and alcohol.

Zhou's lawyer argued that he had no criminal record of graduating from Northeastern University as a Magna Kum Raude (honor) and was not a violent person, having suffered a "very difficult mental health crisis." Contrary to the content of the message, he also explained that he did not actually own a firearm.

사진 확대

OpenAI says that if a user is detected trying to hurt others, a trained small review team can separately examine the conversation and hand it over to law enforcement if it is determined that there is an urgent physical threat to others. However, the company does not report cases of self-harm to the authorities for privacy protection.

This case contrasts with the controversial case of OpenAI's failure to report in the past. In April, the bereaved families of the Tumblidge shooting in British Columbia, Canada, filed a lawsuit against the company and CEO Sam Altman, claiming that OpenAI did not inform the police even though the gunman left a violent conversation on ChatGPT months before the attack. According to NPR reports, OpenAI's automation system had already marked the then 18-year-old criminal's account in June 2025 in relation to gun violence, but OpenAI says the conversation did not meet the law enforcement agency's transfer criteria at the time. Nine people were killed in the incident.