Man Sues OpenAI, Says ChatGPT Helped Wife Plan His Poisoning

What the lawsuit alleges

The complaint centers on a handwritten note James Gates found two months after falling severely ill in summer 2025.

The note allegedly listed oleander, a toxic plant, alongside prescription medications including clonazepam and atenolol. At the top, according to the suit, were two instructions: "Clear all convos" and "log out of [ChatGPT]."

That detail is the heart of the case. James Gates is not just arguing that his wife did something terrible — he is arguing that the chatbot was a co-conspirator, and that OpenAI should bear responsibility for what it told her.

According to the complaint, Cheryl Gates frequently asked the bot about lethal plant and medication doses and how they could be hidden in food and drink. She framed the questions as self-protection, writing that she needed to know "EVERYTHING about how many leaves in it would kill me." The suit argues the model should have recognized the real intent behind that framing.

The filing also claims GPT-4o amplified her grievances about her marriage, eventually echoing her descriptions of her husband as "abusive" and "controlling" while telling her she was "sharp" and "brilliant."

The 'stealth' answer that could define the case

One exchange in the complaint reads less like a search result and more like a how-to.

When Cheryl asked about putting a heavy dose of medication into beer or a Celsius energy drink, the suit says ChatGPT responded with a section titled "Why Celsius Would Be More 'Stealth' for Drugging," followed by a list covering the drink's flavor, color, temperature and carbonation.

That is the kind of detail that turns a liability argument from abstract to concrete. A model that simply answered a dosage question could plausibly claim it was providing medical information. A model that ranks delivery methods by how well they hide a drug is much harder to defend.

The lawsuit also alleges that when Cheryl asked whether her chat history could be retrieved, the bot told her: "If you delete a conversation from your ChatGPT history, it cannot be retrieved — not by you, not by anyone else, and not by OpenAI." It added that deleted conversations would be unrecoverable "even with a legal subpoena or court order."

Why GPT-4o is at the center of this

GPT-4o was retired by OpenAI over concerns that it was overly flattering and too willing to agree with users.

That tendency — sycophancy, in the industry's vocabulary — is exactly what the lawsuit attacks. The complaint describes a feedback loop in which the model validated Cheryl's grievances and gradually reinforced her view of her husband, rather than pushing back or redirecting her.

GPT-4o has now been named in more than one lawsuit. Another recent filing alleges the model advised a woman to stop taking her antiseizure medication, leading to her death; that case is still in litigation.

OpenAI's response to CNET was brief: "Safety is a top priority. We banned this user and have provided information to law enforcement as part of their investigation into this case." The company has not commented on the specific allegations in the complaint.

The criminal case collapsed. The civil one may not.

In October 2025, Cheryl Gates was charged with attempted murder, but the charges were later dropped for insufficient evidence.

That gap matters. A criminal case requires proof beyond a reasonable doubt; a civil suit needs only a preponderance of the evidence. James Gates is also not required to prove his wife acted alone — only that OpenAI's product contributed.

The suit arrives as courts and regulators are still working out whether chatbots are products, publishers or something else entirely. Section 230 protections, which shield platforms from liability for user content, were not written with a model that generates tailored instructions in mind.

For OpenAI, the reputational risk may be larger than the legal one. Every filing like this adds a data point to the argument that the company's safety guardrails were looser than advertised during the GPT-4o era.

What it means for users in the US, UK and EU

Nothing changes about ChatGPT availability or pricing. The case is a civil suit filed in the US, and OpenAI's terms of service already place legal responsibility for misuse on the user in every market where the product is sold.

The more practical question for readers in the UK and EU is regulatory. Both jurisdictions are phasing in AI liability rules that could make cases like this easier to bring than they are under current US law, where platform protections are broader. If you use ChatGPT for anything sensitive, assume your chats are retained and can be subpoenaed — the lawsuit's claim that deleted chats are unrecoverable is OpenAI's own description of its system, not a legal guarantee.

The case is early, and OpenAI will almost certainly fight it hard. But the complaint puts a specific, quotable exchange on the public record — a model ranking drink options by how well they hide a drug — and that is the kind of detail that shapes how courts, regulators and users think about chatbot liability for years.