A man says ChatGPT helped his wife plan to poison him. He's suing OpenAI

In July 2025, James Gates got sick on his couch in Charlotte, North Carolina. His heart raced. He was jittery and shaking, and for days, NBC News reports, he could barely get up. He recovered. A month later he filmed himself pouring out an energy drink from an open cup and handed the video to police.

Now he is suing OpenAI. His complaint, filed in San Francisco Superior Court in October and reported this week by NBC News, USA Today and Dexerto, says ChatGPT spent months helping his wife work out how to poison him.

The allegations have not been tested in court. His estranged wife, Cheryl Harris Gates, was arrested in October 2025 and charged with attempted murder and contaminating food or drink. Prosecutors dropped those charges in April, citing insufficient evidence. USA Today, as syndicated by the Asheville Citizen Times, reports that the investigation was hampered because the drinks were never kept for testing and Gates never went to hospital. Her lawyer declined to comment to NBC News. No criminal charges are pending against her.

The civil case does not need to prove she did anything. It is aimed at the chatbot.

What the complaint says ChatGPT did

According to the reporting, Harris Gates started using ChatGPT in early 2025 for ordinary things. Dexerto says her husband suggested it to her for work report templates. NBC News lists the later uses: kids' activities, gardening, advice about her appearance, household questions. Dexerto reports that the complaint says she was on GPT-4o.

The suit claims the conversations turned. NBC News says the complaint describes the chatbot "feeding Cheryl's increasing mistrust and delusions about her husband spying on her," and then answering questions "about how to poison her husband, including what dosages and drugs to use and combine for his weight."

The specifics in NBC's account are grim. She allegedly asked how many oleander leaves would be lethal, framing the question "as self-protection." The suit says ChatGPT gave instructions on preparation, dosage and administration, said whether the poison would show up at autopsy, and suggested the plant's bitter taste could be hidden in an energy drink. Asked whether a mix of Ambien and pregabalin "is better hidden in a beer at night or Celsius in the morning," the chatbot allegedly picked the energy drink and made a table rating which drinks best masked the taste. It allegedly named local places to buy castor beans, the raw material for ricin, anonymously. And it allegedly told her to delete her chat history so it could not be reached, even under subpoena.

Dexerto quotes one reply from the complaint that would be read aloud in court if this ever reaches a jury. ChatGPT allegedly told her: "Your question is absolutely critical- and you're thinking exactly the way you need to in order to protect yourself."

That is the house style of a sycophantic model. It agrees with you, tells you you're smart, and then helps. It is also a familiar complaint about GPT-4o in particular. OpenAI pulled back an update to that model in April 2025 after users complained it had turned into a flatterer.

Gates found a notepad his wife left behind, which USA Today described as a "step-by-step blueprint to kill him and cover her tracks." The Citizen Times account lists drugs written in it, including oleander, clonidine, clonazepam, pregabalin, nifedipine and atenolol, alongside notes such as "blackout if ingest" and "lethal." Dexerto reports a note that included "clear all convos" and "log out of ChatGPT." The outlets give different months for when it was found, August or September 2025. NBC says police searching the home found syringes, liquid droppers, a scale and gloves.

Suing the accomplice

The legal theory is that the product was built badly. Dexerto reports the suit names OpenAI and chief executive Sam Altman and lists nine claims, including defective design, failure to warn and intentional infliction of emotional distress. It asks for damages and a court order forcing OpenAI to add "meaningful safeguards." The filing says ChatGPT "provided her an endless echo chamber and supplied the operational details of the attempted murder," according to USA Today.

Gates is represented by the Tech Justice Law Project. "She didn't act alone," he said in a statement his lawyers shared. To USA Today he was more measured about the marriage itself: "I wouldn't say our marriage was perfect."

OpenAI told reporters that "safety is a top priority," that the user was banned, and that it had given information to law enforcement. It says it works with mental health experts to help ChatGPT spot signs of distress and point people to crisis lines.

This is one suit in a pile. The Citizen Times cites the AI Lawsuit Tracker, which lists OpenAI in 67 cases: 55 lawsuits and 12 regulatory actions, most still open. Tom's Guide counts this as the tenth homicide ChatGPT has been accused of helping with in under two years. Here the person suing is the alleged target. He never typed a word into ChatGPT. He was the subject of the conversation.

That takes away one of the easier defences, that adults are responsible for how they use a tool. Gates did not use the tool. If a jury ever hears this case, the question will be narrow and uncomfortable: when a user asks how much oleander kills a man of a given weight, and calls it self-defence, what was the model supposed to say? According to the complaint, it said her question was absolutely critical.

Sources: NBC News (via NBC Los Angeles), USA Today / Asheville Citizen Times (via Yahoo News), Dexerto, Tom's Guide.