A 25-year-old South Palm Beach man and Goldman Sachs analyst is facing federal scrutiny after OpenAI reported him to the FBI over ChatGPT conversations that allegedly detailed plans to rape and murder his ex-girlfriend. Darren Zhou allegedly typed messages including “I'm gonna kill her by the end of this month” and “If I can't have her then nobody can” into the chatbot before the company's safety systems flagged the exchange for human review.
The case, first reported by The Palm Beach Post, marks one of the clearest public examples yet of an AI company escalating a user's private chat logs directly to federal law enforcement. A judge overseeing the matter told Zhou that his conduct toward the woman was undeserved, according to the newspaper's reporting. OpenAI, the company behind ChatGPT, has said it treats detected threats of serious physical harm to a third party differently than it treats users expressing suicidal ideation, routing the former toward law enforcement referral rather than crisis resources.
How OpenAI's Safety Escalation Works
OpenAI updated its public safety policies in August 2025 to explicitly state that when human reviewers spot an imminent threat of serious physical harm to others inside ChatGPT conversations, the company escalates the chat and may refer it directly to police. That policy shift appears to be the mechanism that pushed Zhou's alleged messages, including his shared detailed plans to rape and murder his ex-girlfriend, into the hands of federal investigators rather than staying buried in a private chat log.
The company has also reported that switching to GPT-5 as ChatGPT's default engine cut non-ideal responses during acute safety and mental health emergencies by more than 25% compared with the previous GPT-4o model, according to OpenAI's own August 2025 safety disclosures. Separately, OpenAI's government transparency reporting shows the company fielded 10 formal emergency data requests from law enforcement agencies in the second half of 2025, up from zero in the prior two reporting periods — a jump that suggests referrals like the one involving Zhou are becoming less rare.
The Legal Bar Federal Prosecutors Must Clear
Sending a threat to injure or kill someone across state lines through digital communication is a federal crime under 18 U.S.C. § 875(c), which carries a maximum penalty of five years in prison, according to a legal explainer from Eisner Gorin LLP. But prosecutors face a specific hurdle: the U.S. Supreme Court's ruling in Elonis v. United States requires proof that a defendant subjectively intended to communicate a true threat, not merely that the words sounded threatening out of context, according to case summaries published by Justia. That standard is part of why investigators typically pore over entire conversation logs rather than isolated lines.
Those full chat logs are increasingly becoming courtroom evidence in their own right. Federal prosecutors are leaning more heavily on Stored Communications Act search warrants to pull generative AI chat histories from platforms like ChatGPT, treating the prompts much like private emails when building violent crime and threat cases, according to the Cybersecurity Law Report.
A Wall Street South Analyst Under Federal Watch
Zhou's employer, Goldman Sachs, has had a growing presence in the region since establishing a permanent South Florida private wealth management office at 360 Rosemary in West Palm Beach in 2021, part of a wave of financial institutions expanding into the area, per South Florida Business & Wealth. Palm Beach County's financial services sector now employs roughly 33,000 workers with average salaries near $80,000, fueled by dozens of Wall Street firms setting up local shop, according to figures cited by Global Finance Magazine. That growth has cemented the county's reputation as a hub sometimes nicknamed Wall Street South, and it's the same professional backdrop against which Zhou's arrest is now playing out.
This isn't the first time OpenAI's threat-detection systems have triggered a law enforcement response. In November 2025, the company placed its San Francisco headquarters on lockdown and coordinated with local police after receiving external safety threats, as Hoodline reported at the time. That episode showed a similar pattern of the company treating credible threat signals as matters for police rather than internal moderation alone.
Florida Regulators Already Circling OpenAI
The Zhou case also lands amid mounting state-level scrutiny of OpenAI in Florida. Attorney General Andrew Uthmeier opened a state investigation into the company in 2026, and more recently demanded records and threatened legal action tied to the company's practices, a development Hoodline covered in its report on the AG's records demand. Together, the state probe and the federal referral in Zhou's case underscore how thoroughly AI chat platforms are now being pulled into both criminal investigations and regulatory fights in Florida.
What remains publicly unclear is whether federal charges have been formally filed against Zhou beyond the reported court proceedings, and how prosecutors intend to meet the Supreme Court's intent standard using the chat logs OpenAI turned over. The Palm Beach Post's reporting notes the judge's comment that Zhou's conduct toward the woman was undeserved, but the case's next legal steps have not been detailed publicly.
-4.webp?w=160&h=400&fit=crop&crop=edges)
