A breakup can make people do strange things. Some call their best friend at 2 a.m., some write a long message and never send it, while others open ChatGPT and start typing everything they are feeling. For Darren Zhou, a 25-year-old Florida man, those conversations took a deeply disturbing turn, and eventually also became part of an FBI investigation.

According to court records and reports, Zhou began using ChatGPT in March 2026, shortly after his 6-month relationship ended. What initially appeared to be conversations about his ex-girlfriend, her routines, hobbies, social life and his desire to get back together reportedly escalated into increasingly violent thoughts and detailed threats. Investigators later described the messages as more than vague emotional outbursts, pointing instead to a pattern of rehearsal and planning.

Allegedly, when Zhou began talking to the chatbot about detail plans to rape and kill his ex-girlfriend, the chatbot not only flagged the conversation but also alerted the FBI about it. Federal authorities subsequently reviewed around 2-months of Zhou’s messages and contacted local investigators. He was arrested in May. In August, Zhou pleaded guilty to aggravated stalking, written threats to kill and illegal use of a cellphone. He received 8 years of probation rather than a prison sentence, with restrictions including no contact with his ex and participation in a batterer intervention programme. The case is horrifying for obvious reasons. But it also raises a much more everyday question for anyone who has ever typed, ‘My boyfriend did this, what should I do?’ or ‘Why is my girlfriend acting like this?’ into ChatGPT:

The appeal is obvious, AI is available at any hour to talk to you. It does not interrupt you, roll its eyes, tell you that you are overreacting or get bored after hearing the same breakup story for the 5th time. When emotions are running high, having something that can immediately help organise your thoughts can feel incredibly comforting. It can also be useful for relatively harmless things: helping you phrase a difficult conversation, suggesting questions to ask your partner, helping you identify patterns in an argument or encouraging you to pause before sending an angry text. But there is an important distinction between using AI as a thinking tool and treating it as your relationship therapist.

AI only knows the version of the relationship that you give it. If you tell ChatGPT that your partner ignored you for 3 hours, it does not automatically know that you had a fight the night before, that they were at work, or that this behaviour is completely unlike them. Give it a one-sided account and you can end up receiving advice built on a one-sided picture.

If ChatGPT becomes the first place you go after every fight, every confusing text and every relationship crisis, you may slowly start outsourcing emotional decisions to a machine. Instead of asking, ‘What do I actually want?’ you can end up asking, ‘What does ChatGPT think I should do?’

And then there is privacy. OpenAI says ChatGPT uses automated systems and human review to enforce its policies, and that conversations indicating an imminent and credible risk of harm to others may be referred to law enforcement. Its current policy also says user data can be disclosed in response to valid legal process, with an emergency exception involving danger of death or serious physical injury. That does not mean that telling ChatGPT your boyfriend forgot your anniversary is going to result in an FBI call. Zhou’s case involved alleged threats of serious real-world violence, which is an entirely different category.

Still, the lesson is worth remembering. Use AI to help you think, not to replace the people who actually know you. For relationship problems, that could mean talking to a trusted friend, having an honest conversation with your partner or, when things become serious or emotionally overwhelming, speaking to a qualified therapist. Because ChatGPT can help you find the words, but it cannot live your relationship for you.