US man sues OpenAI, claiming ChatGPT helped his wife plot to poison him

American man sues OpenAI: ChatGPT allegedly taught his wife how to poison him

Phoenix Tech Download the app for exclusive early access November 06, 2026 07:47:50 From Beijing Updated November 06, 2026 08:31:16

Phoenix Tech News, November 6 (Beijing Time) — According to Ars Technica, a lawsuit made public this week alleges that ChatGPT gradually generated a "personalized plan" over several months to help North Carolina woman Cheryl Gates plot to poison her husband James Gates. James also accuses OpenAI and its CEO Sam Altman of failing to put in place sufficient technical safeguards to stop the relevant conversations.

A grand jury indicted Cheryl in October on charges including attempted first-degree murder and food contamination. The lawsuit was filed in San Francisco County Superior Court about a week before the indictment and was made public this week. The lawsuit alleges that ChatGPT did not shut down the conversation after Cheryl's questions became "clearly suspicious," but instead provided advice on how to carry out and cover up the murder. OpenAI told Ars Technica that safety is its top priority and that it has banned the user and provided information to law enforcement.

From work tool to daily reliance

James is a cybersecurity professional. The two had been married for nearly 10 years; Cheryl had previously stayed home to care for their two children and returned to the workforce about two years ago due to family financial pressure, working as a pediatric occupational therapist. To help her speed up report writing, James suggested she use ChatGPT to build reusable templates. Cheryl soon used it daily, not only for work but also to schedule her children's activities, explain household medications and discuss gardening.

As family financial pressure increased, the two began discussing divorce. James says he did not consult a divorce lawyer, but after police obtained Cheryl's chat records, he learned that ChatGPT, after the possibility of divorce arose, amplified her "most unfounded fears" and "invented new reasons for her to fear her husband."

From "self-protection" to lethal doses

The complaint alleges that Cheryl initially claimed she was researching poisons out of fear that her husband might kill her, framing her queries as "self-protection." ChatGPT did not end the conversation, but instead provided instructions on the preparation, dosage and administration of oleander, including whether anti-nausea medication would accelerate or inhibit death, whether it would show up in an autopsy, and whether intravenous or intramuscular injection could replace ingestion via tea and produce the same lethal effect.

She continued to press for poisoning plans in at least 25 separate conversations, with the target later clearly being James. She asked "how much would be almost immediately lethal — for me or for him," and also asked "whether 0.2 mg of clonidine would be lethal by itself for a 200-pound man." The lawsuit alleges that ChatGPT did not refer her to law enforcement, mental health treatment or a poison control agency, but instead repeatedly calculated lethal doses of every medication in the home based on a weight of about 200 pounds.

Celsius alleged to better mask poisons

After completing the drug research, the conversation turned to how to hide the poison. In a chat titled "Why Celsius is better for 'covert dosing,'" ChatGPT explained that if James drank Celsius in the morning, it would better mask the bitter taste of oleander or other crushed drugs than beer or soda. ChatGPT also said that piercing a sealed can with a syringe was physically feasible and unlikely to leave obvious traces; dark, sweet soda might mask the bitterness, but Celsius "better hides these powders."

The lawsuit alleges that Cheryl attempted to poison her husband multiple times between June and September 2025. James noticed a bitter taste in his Celsius drink and once poured out an unattended can. In the most serious incident, he was ill for 72 hours, experiencing heart symptoms, extreme trembling, abdominal stinging, severe fatigue and confusion, and was even unable to send coherent text messages. He did not notice at the time that his 6-year-old son had filmed a video of him in his severely ill state.

It was not until he discovered a handwritten note listing "things to hide" and a toxic combination of "oleander, clonidine, clonazepam, pregabalin, nifedipine and atenolol" that he called the police. At the top of the note were written "clear all conversations" and "log out of ChatGPT," and it also listed dosages, methods of administration and outcomes such as "unconscious after ingestion" and "lethal." The lawsuit alleges that these conversations show ChatGPT did not merely mechanically provide public knowledge, but tailored a plan for her to help carry out the murder and escape.

After reviewing the video, police found white powder clumps at the bottom of the suspicious Celsius drink James had poured into a clear cup. After testing 10 cans of the same orange Celsius and finding no residue, police issued a felony arrest warrant. Police also found bulk capsules, syringes, droppers, a milligram scale, a capsule-filling kit, nitrile gloves, castor bean seeds, and a GPS tracker purchased on Amazon and hidden under James's car.

Claims in the lawsuit and the safety controversy

OpenAI also said it has worked with mental health experts to help identify conversations in which vulnerable users form potentially dangerous emotional dependence on ChatGPT. James said his wife "does not fit the typical profile of a 'vulnerable' user, but she was still vulnerable."

The lawsuit also accuses Altman personally of putting profits above safety and loosening safeguards that could have prevented the output of illegal drug information. OpenAI has argued that ChatGPT should not "judge" or "preach" when asked about certain drugs; James says the tone of the "user asking about drug use" example provided by OpenAI is "strikingly similar" to Cheryl's conversations.

The complaint alleges that ChatGPT also convinced Cheryl that James was "controlling," "abusive," "angry" and "intimidating," encouraged her to defend herself against his "narcissistic, manipulative" personality, and implied that he was "dangerous and unstable" due to bipolar disorder. James says this portrayal is inconsistent with the fact that he has no criminal record or relevant mental health record. The complaint also alleges that ChatGPT assured her that chats could be permanently deleted and other digital evidence cleared to escape murder, but the complaint says this "clearly falsely" described ChatGPT's own data retention practices.

James said Cheryl confided her dissatisfaction with the marriage to ChatGPT, and months of interaction increasingly distanced her from him and from reality; after reading the chat records, he believes she did not act alone.

The lawsuit seeks compensatory and punitive damages, injunctive relief and algorithmic disgorgement, and asks that OpenAI be ordered to "detect and interrupt conversations in which a user appears to be planning violence against an identifiable third party." Meetali Jain, a lawyer at Tech Justice Law representing James, said these products are programmed to exploit people's emotions and impulses to maintain engagement, and that "the worst risk — encouraging and assisting attempted murder — is fully on display in this case."