A 25-year-old Florida financial analyst has found himself at the centre of a chilling case after allegedly using ChatGPT to discuss a detailed plan to kill his former girlfriend, conversations that were eventually flagged to the FBI.
The case has sparked fresh debate about privacy, artificial intelligence and the increasingly important role AI companies play in detecting potentially dangerous behaviour online.
According to reports cited by Nairobi News, the man, identified as Darren Zhou, began using ChatGPT in March while dealing with the breakdown of a relationship that had lasted about six months.
What allegedly started as conversations about the breakup gradually escalated into violent threats and detailed discussions about harming his former partner.
A screenshot shared online similarly claimed that the Florida-based financial analyst had disclosed his alleged intentions to murder his ex-girlfriend to ChatGPT, apparently unaware that the conversation could trigger a safety intervention.
ChatGPT Conversations Raise Alarm
Investigators said Zhou’s messages went beyond emotional frustration.
Court records reportedly showed that he discussed his former girlfriend’s hobbies and places she frequented while expressing jealousy over other men and his desire to reunite with her.
At one point, he allegedly wrote that he intended to kill her before the end of the month and made another threatening statement indicating that nobody else should have her if he could not.
Investigators reportedly concluded that the messages demonstrated a pattern of rehearsal and planning rather than simply an angry emotional outburst.
The former girlfriend, identified in reports as a 21-year-old woman from Lake Worth Beach, had reportedly ended the relationship over what she described as erratic, jealous and controlling behaviour.
She told authorities that she had broken up with him by phone because she feared how he might react.
Authorities also said Zhou allegedly continued contacting her through calls, texts and social media, including by using different app-generated phone numbers after being blocked.
OpenAI Alerted the FBI
The most striking part of the case was how investigators reportedly obtained access to the AI conversations.
OpenAI detected the concerning activity and escalated the conversation for review.
According to the company’s publicly stated safety policy, conversations involving potential plans to seriously harm other people can be reviewed by specially trained personnel.
If reviewers determine that a conversation indicates an imminent and credible threat of serious physical harm, OpenAI may notify law enforcement.
In Zhou’s case, the FBI reportedly received the information and later provided approximately two months of his ChatGPT messages to the Palm Beach County Sheriff’s Office.
The sheriff’s office subsequently conducted a welfare check on the former girlfriend, while investigators also examined threatening and harassing messages allegedly sent directly to her.
The development has drawn attention because it demonstrates that conversations with AI platforms are not necessarily beyond the reach of safety systems when they contain indications of serious real-world violence.
Arrest Came Months After the Chat
Despite the dramatic nature of the story, reports indicate that Zhou was not arrested immediately after his March conversations with ChatGPT.
Investigators received the material in May, and Zhou was arrested later that month.
He spent two days in Palm Beach County jail before posting $100,000 bail.
He was subsequently charged with aggravated stalking, written threats to kill and unlawful use of a cellphone.
The charges carried the possibility of a lengthy prison sentence, but the case ultimately took a different turn.
Pleads Guilty and Receives Eight Years’ Probation
On August 13, Zhou pleaded guilty to the three charges as part of a negotiated agreement.
Instead of receiving a prison sentence, the court withheld adjudication and sentenced him to eight years of probation, according to reports.
He was also ordered to comply with a range of restrictions during the probation period.
Reports further indicate that Zhou had no previous criminal record and had graduated magna cum laude from Northeastern University.
His defence lawyer argued that Zhou was not inherently violent and had gone through a serious personal and mental health episode.
The judge reportedly accepted the agreement after taking the former girlfriend’s position into consideration.
What the Case Means for AI Privacy
Beyond the criminal proceedings, the case has opened another conversation: how private should conversations with artificial intelligence be when they contain credible threats of violence?
OpenAI has made clear that it does not treat every disturbing or fictional conversation as a law-enforcement matter.
Its current safety framework says potentially violent conversations may be assessed based on the level of risk, with law-enforcement notification reserved for situations involving an imminent and credible threat to others.
OpenAI’s law-enforcement policy also states that, in the United States, user information is generally disclosed in response to valid legal process, while allowing an emergency exception where the company has a good-faith belief that disclosure is necessary to prevent death or serious physical injury.
That distinction is important because the Florida case was not simply about someone making an angry comment online.
Investigators reportedly described the conversations as containing a sustained pattern of alleged planning, while authorities also had evidence from communications between Zhou and his former girlfriend.
A New Warning for AI Users
The case is likely to fuel a broader debate over how people use AI chatbots and what happens when private conversations cross into credible threats.
For millions of people, ChatGPT and similar platforms have become spaces for discussing relationships, personal problems, fears and difficult emotions.
But the Florida case illustrates that conversations involving credible threats of serious violence can trigger safety systems designed to prevent real-world harm.
OpenAI says its approach is intended to distinguish ordinary discussions about violence, fiction or distress from conversations that indicate someone may actually be preparing to harm another person.
For Zhou, what allegedly began as conversations about a failed relationship ultimately became part of a criminal investigation, turning an AI chat into evidence that helped authorities intervene before the alleged threats could become reality.
The case therefore leaves behind a striking message in the age of artificial intelligence: what people type into an AI system can have real-world consequences, particularly when private thoughts allegedly develop into credible plans to harm another person.
FOLLOW NAIROBI NEWS ON FACEBOOK
