MEDIANAMA
Technology and policy in India
A 25-year-old former Goldman Sachs financial analyst pleaded guilty on August 13 to felony stalking and threat charges after OpenAI flagged months of ChatGPT conversations in which he described plans to harm his ex-girlfriend and notified the FBI, according to a report by The Detroit News.
Darren Zhou, of South Palm Beach, Florida, received eight years of probation and two years of electronic monitoring under a plea deal that avoided prison time and a felony conviction. Circuit Judge Scott Suskauer said he accepted the agreement because Zhou’s ex-girlfriend, a 21-year-old from Lake Worth Beach, approved it.
The charges and the deal: Assistant State Attorney Ana Cuskova charged Zhou in June with aggravated stalking, written threats to kill, and illegal use of a cell phone. The felonies carried a combined maximum sentence of 25 years in prison. Zhou pleaded guilty to all three counts. Because the judge withheld formal adjudication, Zhou will avoid a felony record if he completes probation.
Zhou spent two days in Palm Beach County jail after his arrest in May before posting $100,000 bail. He was later terminated from his position at Goldman Sachs.
How did OpenAI catch it? According to OpenAI’s stated policies, a small internal team reviews ChatGPT conversations when the system detects that a user may be planning to harm others. In serious cases, the company forwards a tip to law enforcement. That is what happened with Zhou’s account: OpenAI detected a pattern in his messages, reviewed the conversations, and alerted the FBI in May 2026.
Federal agents provided approximately two months of Zhou’s chat logs to the Palm Beach County Sheriff’s Office. A sheriff’s deputy who reviewed the material, which included only Zhou’s messages and not ChatGPT’s responses, said the messages were not vague venting but demonstrated a clear pattern of rehearsal and planning.
What did Zhou tell the chatbot? Court records show that Zhou repeatedly told ChatGPT he intended to kill his ex-girlfriend. “I’m gonna kill her by the end of this month,” he wrote in one exchange, adding that if he couldn’t have her, no one could.
In subsequent messages, he outlined a scenario in which he would wait for her with flowers in the gym parking lot and, if rejected, planned to shoot her before taking his own life. Additional messages contained graphic descriptions of sexual assault and torture, including plans to hold her at gunpoint, force her to drive to his home, and shoot her as well as any responding police officers.
Zhou also told the chatbot that he had purchased an AR-15-style rifle, a Glock handgun, and a 12-gauge shotgun, and claimed to have sent his ex-girlfriend photos of the weapons along with zip ties and latex gloves. His attorney later said that Zhou did not actually possess any firearms. He also documented her routines, hobbies, and frequented locations, and expressed jealousy toward other men in her life.
The harassment outside the app: According to investigators, while Zhou shared his plans with ChatGPT, he also contacted his ex-girlfriend using multiple app-generated phone numbers. She documented messages that were harassing, insulting, sexual, and threatening, including one sent on her 21st birthday and another that named the gym she was visiting at the time.
She later told deputies that lingering feelings for Zhou and the hope that ignoring him would end the harassment delayed her report. By the time the FBI alerted local investigators, Zhou had expanded his threats to include her family. Deputies conducted a welfare check before arresting Zhou, using his messages to both ChatGPT and his ex-girlfriend as evidence.
Part of a wider pattern: This case joins a small but increasing number of cases in which AI chatbot logs have been used as evidence in criminal investigations. However, it is less common for OpenAI to proactively alert police than for logs to be subpoenaed after the fact.
This contrasts with an earlier case in which OpenAI did not alert authorities. In February, OpenAI informed Canadian officials that it had banned a ChatGPT account belonging to Jesse Van Rootselaar months before he killed eight people in Tumbler Ridge, British Columbia. The account was banned because of concerns about violent activity, but police were not notified because the conversations did not indicate an imminent threat. The disclosure prompted anger from Canadian officials, and Artificial Intelligence Minister Evan Solomon called on OpenAI leadership to review its security protocols. OpenAI has since stated that its current protocols would require police notification in a similar situation.
Researchers have also expressed concern about frequent chatbot use during mental health crises, a phenomenon some refer to as “AI psychosis.” In certain cases, this has preceded involuntary commitment or arrest.
Regulators in other regions have started addressing these risks, including AI psychosis, directly. In China, new rules that took effect on July 15 require companies offering AI companion apps to remove features that encourage emotional attachment. Issued by the Cyberspace Administration of China and four other agencies, the measures prohibit romantic or family-style virtual relationships with users under 18, mandate that apps notify a user’s emergency contact or guardian if signs of a mental health crisis are detected, and require government safety reviews. Officials have linked the crackdown in part to concerns that strong emotional bonds with chatbots could further reduce China’s birth rate, which reached a record low in 2025.
Also read:
For You
- Read Reasoned by Nikhil Pahwa: How AI is changing our world
- Sign up for MediaNama’s Daily Newsletter to receive regular updates
- Sponsor a MediaNama Event
Elon Musk says government censorship on X is now visible, but India’s Section 69A framework keeps government blocking orders confidential.
Indian ministries are reportedly setting up Quick Response Teams to monitor social media, flag fake or misleading content, and coordinate with the PIB Fact Check Unit to issue fact-checked responses.
MediaNama is the premier source of information and analysis on Technology Policy in India. More about MediaNama, and contact information, here.
© 2024 Mixed Bag Media Pvt. Ltd.
source