When a ChatGPT Conversation Reached the FBI

Detik Republik | Florida - A Florida case highlights the limits of privacy when people turn to AI for deeply personal conversations

A case in Florida in 2026 offers a striking answer—and raises difficult questions about how private our conversations with AI really are.

In March 2026, Darren Zhou, a 25-year-old financial analyst at Goldman Sachs, began using ChatGPT after the end of a six-month relationship.

At first, he discussed the breakup, his former girlfriend and his desire to reconcile.

The conversations soon became much darker.

According to court records, Zhou repeatedly described plans to rape and kill his former girlfriend. In one message, he wrote: “If I can't have her then nobody can.”

He also described a plan to wait for her near a gym, confront her with flowers and, if she rejected him, shoot her before killing himself.

The messages continued into April, becoming increasingly specific. Investigators later described the exchanges as a pattern of planning rather than isolated emotional outbursts.

In May, OpenAI flagged the conversations and reported the case to the FBI.

The FBI then passed information from the case to the Palm Beach County Sheriff's Office in Florida. According to investigators, the material included roughly two months of Zhou's ChatGPT messages.

Local deputies conducted a welfare check on Zhou's former girlfriend and began investigating his conduct outside the AI platform as well.

Investigators found that Zhou had allegedly continued contacting her through calls, text messages and social media after the breakup.

Armed with the information from both sources, deputies arrested Zhou in May 2026.

He spent two days in jail before posting $100,000 bail.

In June, prosecutors formally charged Zhou with three felonies: aggravated stalking, written threats to kill or cause bodily harm, and illegal use of a communications device to facilitate a felony.

The charges carried a potential prison sentence of up to 25 years.

Goldman Sachs subsequently ended its relationship with Zhou.

On August 13, Zhou pleaded guilty to all three charges.

Under a negotiated agreement accepted by the court, he avoided a prison sentence and a formal felony conviction. The judge withheld adjudication and sentenced him to eight years of probation, including two years of electronic monitoring.

The judge said he accepted the agreement after learning that Zhou's former girlfriend supported it.

The timeline matters because the case demonstrates something often misunderstood about AI privacy.

The messages were not simply “sent to the police” the moment Zhou typed a threatening statement.

There was a sequence:

March–April: Zhou sends increasingly specific threats to ChatGPT.

May: OpenAI flags the conversations and alerts the FBI.

May: The FBI refers the information to Palm Beach County authorities.

May: Zhou is arrested.

June: Prosecutors file felony charges.

August: Zhou pleads guilty and receives probation.

The case therefore does not show that every ChatGPT conversation is routinely read by humans.

OpenAI has said it uses automated systems to identify potentially harmful activity, with human review in certain cases. Its policies also provide for contacting law enforcement in situations involving credible and imminent threats of serious physical harm.

That is a very different proposition from saying that employees routinely monitor everyone's conversations.

But it does mean that users should not confuse a private-feeling conversation with an absolutely private one.

People increasingly use AI to discuss matters they once might have shared only with friends, family or professionals.

They talk about relationships, financial problems, workplace disputes, health concerns and personal fears. Some upload sensitive documents.

AI can feel remarkably intimate because it responds immediately, remembers context in some settings and does not appear to judge.

But the technology remains a digital service.

A conversation can feel private while still being subject to the platform's data practices, safety systems and applicable law.

That distinction is becoming increasingly important as AI moves deeper into everyday life.

The Florida case is therefore about more than one man, one chatbot and one criminal investigation.

It raises a question that millions of AI users may eventually have to confront: