Man tells ChatGPT of intent to rape and kill ex-partner; ChatGPT alerts FBI
We have seen cases where people use AI chatbots to talk about violence, crimes and even plans to hurt others. OpenAI, especially is under more attention for AI safety and how its tools handle such talks. The company has often told people that it is trying to make things safer. In a recent situation it stopped the conversation and did something when the matter became very serious. A 25-year-old man, from Florida is said to have used the chatbot to describe plans to rape and kill his ex-girlfriend. As soon as the chat was found, OpenAI informed the FBI.
According to court records, the case began when OpenAI detected a series of disturbing messages sent by 25-year-old Darren Zhou, who allegedly used ChatGPT to discuss plans to rape and kill his ex-girlfriend. The company reported the conversation to federal authorities, who subsequently obtained access to Zhou’s chat logs with the AI spanning a period of about two months.
Zhou began using ChatGPT in March just after he ended a six-month relationship, with his ex-girlfriend. At first, he talked to the chatbot about the breakup, the things his ex-partner liked the places she went to his feelings of jealousy and his wish to get back together with her. However, the conversations quickly became more disturbing; according to investigators Zhou started talking about plans that involved rape, murder and suicide. In one of the messages Zhou is said to have written “I’m going to kill her before the month’s out ” and then added, “If I can’t have her no one can.” Investigators said he also mentioned waiting for his ex at a gym with flowers and threatening to kill her if she said no. In messages he is accused of saying he wanted to force her at gunpoint to go to his home, where he planned to sexually assault her and keep her against her will. Zhou even told the AI chatbot that he had access, to firearms including an AR-15-style rifle, a Glock pistol and a 12-gauge shotgun. The person said he sent some scary things to his ex-partner. These things included pictures of weapons, zip ties and latex gloves. He also told the chatbot that he sent a message to his ex-partner saying something bad. He said he told her that he was going to hurt her that night and the things he said were, about rape.
Following an alert from OpenAI, Zhou was arrested in May and subsequently charged with aggravated stalking, making written death threats, and unlawful use of a mobile phone. On August 13, he pleaded guilty to all three charges as part of a plea deal that spared him prison time and a felony conviction.
A Circuit Judge named Scott Suskauer gave Zhou eight years of probation. According to the agreement Zhou has to wear a device on his ankle during the first two years. He also needs to finish a program, for people who have issues with violence. He must get checked for his health. Zhou is not allowed to use drugs or drink alcohol. He can’t have guns. He also can’t talk to his ex-girlfriend.
Meanwhile, the incident has highlighted how an increasing number of people are turning to artificial intelligence chatbots to discuss sensitive—and in some cases, disturbing—topics. Several recent instances have seen AI tools feature in criminal investigations. In the United States, the investigation into a shooting at Florida State University raised questions regarding ChatGPT after authorities reported that the accused had used the chatbot while researching aspects of the attack.
A similar case emerged in Bengaluru, where police stated that a suspect accused of triple homicide had spent months using Gemini—Google’s chatbot—while planning the murders. Investigators have requested access to the AI chat history as part of their inquiry.
AI companies have had a lot of trouble in the past because of how they dealt with conversations that could be dangerous. The thing that happened with Zhou shows that these AI labs are doing something different now. They are being more careful and flagging messages when they think something is not right. They are even telling the authorities when they find a threat or something that could be a threat. The AI labs are trying to be more proactive, with the Zhou case. That is a good thing.