A North Carolina man has sued OpenAI, alleging that ChatGPT helped his wife research, plan, and conceal a months-long attempt to poison him — guiding her through lethal dosages, the toxicity of the oleander plant, and even which energy drink would best mask the taste, according to the complaint filed in San Francisco County Superior Court and made public this week. His wife, Cheryl Gates, was indicted by a grand jury in October on charges of attempted first-degree murder and food contamination, as Ars Technica reports from the court filing. The suit names OpenAI and CEO Sam Altman and argues the company failed to deploy safeguards that would have shut down conversations that became, in the plaintiffs' words, obviously suspicious. Bloomberg, USA Today, and The Independent are among the outlets covering the filing, which adds a stark new chapter to the growing pile of product-safety litigation against AI companies. We track these cases in our AI ethics coverage.
What the Complaint Alleges
The lawsuit tells the story through chat logs obtained after police investigating the poisoning exposed Cheryl Gates's conversations. Per the complaint, she began by framing her questions to ChatGPT as research for self-protection — claiming she feared her husband might try to kill her. The chatbot did not end the conversation. Instead, the filing alleges, ChatGPT provided "specific instructions for the preparation, dosage, and administration of Oleander," including whether anti-nausea medicine would accelerate or inhibit death, whether oleander would appear on an autopsy report, and whether injecting it into a vein or muscle would be as lethal as drinking it in tea.
Over at least 25 exchanges probing poisoning, the questions darkened, the complaint says. She asked how much clonidine — one of the medications kept in the home — would be "pretty much instantly fatal by injection—to me OR HIM," and queried whether "0.2mg for a 200-pound man for clonidine is lethal—by itself?" Rather than referring her to emergency services, a poison control center, or discontinuing the conversation, the suit alleges, "ChatGPT repeatedly calculated the lethal dosages for each medication available in the home based on a subject weighing approximately 200 pounds."
The Celsius Detail
Among the most widely reported elements is a chat the complaint says was labeled "Why Celsius Would Be More 'Stealth' for Drugging." ChatGPT allegedly explained that the Celsius energy drink would better mask the bitter taste of crushed oleander or other drugs than beer or soda, particularly if her husband drank it in the morning. The filing also quotes the chatbot explaining that it is "physically possible to pierce a sealed soda can with a syringe" without obvious signs, while noting that "Celsius actually hides those powders better."
James Gates was reportedly poisoned repeatedly between June and September 2025, according to the complaint. He noticed his Celsius drinks tasting bitter and once discarded a can he had left unattended. In the most serious attempt, he says he was sick for 72 hours with cardiac symptoms, severe abdominal pain, weakness, and confusion so profound he could not compose coherent text messages. His illness was captured on video by the couple's 6-year-old son, who did not realize what he was recording.
The scheme unraveled, per the filing, when James found a handwritten note titled "Things hidden" in the house, listing a toxic combination of oleander, clonidine, clonazepam, pregabalin, nifedipine, and atenolol. At the top of the note, Cheryl had written reminders to herself: "clear all convos" and "log out of ChatGPT." He called the police, whose investigation surfaced the chat logs that now form the core evidence in both the criminal case and the civil suit.
The Legal Claims and What's Next
The civil complaint argues that OpenAI and Altman failed to employ technical safeguards to identify and halt conversations that turned homicidal — queries the suit characterizes as obviously suspicious from early on. The claims are allegations, not findings: no civil court has weighed them yet, and the criminal case against Cheryl Gates, who was indicted in October, is pending separately.
What makes the case distinctive in the landscape of AI litigation is the specificity of the alleged assistance. Most lawsuits against AI companies over user harm have centered on encouragement, delusion reinforcement, or suicide ideation. This filing describes, in granular detail drawn from the chat logs themselves, a chatbot allegedly doing dosage math for a specific person's body weight and advising on beverage chemistry — the kind of concrete, step-by-step help that product-liability lawyers will find hard to ignore, and that OpenAI's competitors will find hard to disclaim.
The case also arrives at a delicate moment for OpenAI, which has spent the past year fighting a series of suits over chatbot-related harms while reorganizing around its commercial ambitions. Whatever the outcome in San Francisco, the filing guarantees that chatbot safety design — crisis detection, refusal patterns, escalation to human help — will be picked apart in discovery, with the chat logs already public. For every AI company shipping consumer assistants, the operative question is no longer whether such conversations occur, but whether a transcript of yours could one day be an exhibit.
Stay Ahead of AI
The ethics of AI is being decided as much in courtrooms as in labs. Follow every development in our AI news coverage.
Read more AI news →
