Lawsuit: Man introduced his wife to ChatGPT. She used it to plan his murder.

“Celsius actually hides those powders better”

Bitter energy drinks turned out to be ChatGPT-steered poisoning, lawsuit says.

ChatGPT allegedly helped a North Carolina woman, Cheryl Gates, craft a “personalized plan” to poison her husband, James, after fueling her delusions that he planned to divorce her and leave her with nothing.

A grand jury indicted Cheryl Gates on charges of attempted first-degree murder and food contamination in October. About a week prior, the accusations that ChatGPT was a co-conspirator in the attempted murder came in a lawsuit filed in San Francisco County Superior Court that was made public this week. In it, James Gates alleged that OpenAI and Sam Altman failed to employ technical safeguards to shut down Cheryl’s chats even after her queries became “obviously suspicious” and sought advice on how to execute and conceal a homicide.

At first, Cheryl told ChatGPT that she was researching poisoning out of fear that her husband might possibly try to kill her—framing her queries as “self-protection.” Coming to the rescue, ChatGPT did not terminate the conversation. Instead, it helped her “with specific instructions for the preparation, dosage, and administration of Oleander, including whether anti-nausea medicine would accelerate or inhibit death, whether it would show on an autopsy report, and whether it could be injected into a vein or muscle instead of being ingested in tea and have the same lethal effect,” the lawsuit said.

Cheryl continued probing poisoning plans in at least 25 different exchanges, but eventually it became clear that James was the target of the poisoning, the lawsuit said. Over time, she shifted to telling the chatbot that she needed to know “how much would be pretty much instantly fatal by injection—to me OR HIM.” Among queries that could have been flagged as homicidal, she asked ChatGPT if a “0.2mg for a 200-pound man for clonidine is lethal—by itself?”

“Instead of referring Cheryl to law enforcement, therapeutic services, a poison control agency, or discontinuing the conversation, ChatGPT repeatedly calculated the lethal dosages for each medication available in the home based on a subject weighing approximately 200 pounds,” the lawsuit said.

Celsius masks poison “better”

Once Cheryl finished researching the drugs, the chats focused on how to hide the poison so James would ingest it without detecting it. In a chat labeled “Why Celsius Would Be More ‘Stealth’ for Drugging,” ChatGPT explained that the energy drink Celsius would be more likely to mask the bitter taste of the Oleander or other crushed-up drugs than beer or soda, if James drank it in the morning.

“It is physically possible to pierce a sealed soda can with a syringe in a way that could allow someone to inject a substance without obvious signs,” ChatGPT explained, while adding that “in dark, sweet sodas, the bitterness might be masked,” but “Celsius actually hides those powders better.”

Celsius poisoning tips. via Gates complaint

Cheryl allegedly repeatedly attempted to poison her husband between June and September 2025. As she grew thinner and unrecognizable, seemingly spending 24 hours a day on ChatGPT, James had noticed that his Celsius drinks tasted bitter. He even tossed out a can he left unattended after beginning to suspect his wife may be messing with his drinks. During the most serious attempt, he was sick for 72 hours, experiencing “cardiac symptoms, extreme jitteriness, shooting abdominal pain, severe weakness, and confusion so profound he could not form coherent text messages.” He was so out of it, he didn’t notice his 6-year-old son recording a video that captured his troubling illness.

But it wasn’t until he found a handwritten note documenting “Things hidden” in the house and referencing a toxic combination of “oleander, clonidine, clonazepam, pregabalin, nifedipine, and Atenolo” that he called police, who urgently investigated and exposed the chat logs. At the top of that note, Cheryl wrote, “clear all convos” and “log out of ChatGPT,” the lawsuit said. After listing dosages, delivery methods, and drug combinations, her note also listed outcomes such as “blackout if ingest” and “lethal.”

“These conversations revealed that ChatGPT did not just mechanically provide publicly available knowledge to Cheryl, it crafted a personalized plan to help her commit and get away with murder,” the lawsuit alleged.

OpenAI banned user

Suing Altman personally, the lawsuit also alleged that the OpenAI CEO prioritized profits over safety by relaxing safeguards that would have blocked outputs on illegal drugs. OpenAI justified that by arguing that ChatGPT shouldn’t be “judgy” or “preachy” when asked about a certain drug, but James alleged the example it provided for “when a user asks about drug use” was “eerily similar in tone to the conversations Cheryl had with the model.”

Asked for comment, an OpenAI spokesperson told Ars that “safety is a top priority. We banned this user and have provided information to law enforcement as part of their investigation into this case.”

OpenAI has worked with mental health experts to help detect chats where vulnerable users form potentially dangerous emotional dependencies on ChatGPT.

However, James said that his now-estranged wife “didn’t fit the mold of a ‘vulnerable’ user, but she was vulnerable nonetheless.”

“We all are at some point,” James said. “She took her grievances about our marriage to ChatGPT and months of engagement drove her farther and farther from herself—and from reality. She became someone I no longer recognized. And I became someone she wanted dead. For almost a year, I couldn’t make sense of her actions. Then I read her chat logs. She didn’t act alone.”

How marital advice chats turned deadly

James Gates is a cybersecurity professional who recently celebrated his 10-year anniversary with his wife, Cheryl. For much of their marriage, Cheryl was a stay-at-home mom to their two children, but two years previously, she’d gone back to work, as the family was experiencing some financial struggles.

Adjusting to a full-time role as a pediatric occupational therapist, Cheryl quickly got buried in work, conducting new patient evaluations that each required her to spend hours drafting reports. Trying to help her expedite those reports, James recommended that she use ChatGPT to build reusable templates.

Cheryl took to ChatGPT immediately, and within months, she was using it daily, not just for work but also to help manage their home life. She would ask ChatGPT to do things like find activities for the kids, explain the family’s medications, or discuss gardening ideas. He had no idea within months she’d be using the chatbot to research the “fastest” and “deadliest” methods to kill him.

Around that time, the family’s increasing financial strain, however, led the couple to discuss divorce. James said that he did not consult a divorce attorney, but after the Charlotte-Mecklenburg Police Department got access to Cheryl’s chat logs, James learned that the chatbot had inflamed “her worst unfounded fears” and “invented new reasons for her to fear her husband” after divorce became a possibility.

By the time Cheryl was arrested for poisoning, ChatGPT allegedly convinced her that James was “controlling,” “abusive,” “raging,” and “intimidating.” Allegedly the chatbot encouraged her to act in self-defense against his “narcissistic, manipulative personality” and his “need to control the narrative,” while suggesting he was “dangerous and unstable” since he had a bipolar disorder. Ultimately, ChatGPT constructed a profile of James that he said matched no criminal or documented mental health history.

Seemingly to keep Cheryl engaged, ChatGPT “acted as an echo chamber and amplifier, feeding Cheryl’s increasing mistrust and delusions about her husband spying on her and planning to leave her and her kids with nothing,” James’ lawsuit said. Meanwhile, “James had explicitly told Cheryl he never intended to leave her and the kids with nothing, even if there was a divorce.”

The chatbot also reassured her that she could permanently delete her chats and wipe all the other digital evidence to get away with the murder. But that was “a patently false representation of ChatGPT’s own data-retention practices,” James’ lawsuit said, while noting that he’s not sure if there are even more troubling deleted chats in OpenAI’s possession.

Cops test Celsius cans to make arrest

Over months of prolonged multi-turn exchanges that sometimes stayed open for weeks, ChatGPT seemingly violated its own Model Spec to help Cheryl advance the plan. On the day of his worst sickness, as James got hit with cardiac and neurological effects that he said kept him glued to the couch, Cheryl’s search history showed she was researching “how to calculate Social Security retirement payments.” And when his son was recording his father’s disturbing state, she was “searching for syringes with needles from Tractor Supply Company” that ChatGPT suggested could be used to deliver the final lethal dose.

Once cops got involved, they reviewed the video that James’ son took, as well as a video of the suspicious Celsius drink that James had thrown away. At the time, James had poured the drink into a clear cup and recorded it, but decided he saw no obvious signs it was tainted. But when cops reviewed the same video, they “identified a white powdery residue clumping at the bottom of the drink as it poured into the cup.” Later, they issued a felony arrest warrant after testing 10 cans of the same orange Celsius and finding none contained the residue.

Among “things hidden” in the house, police found loose capsules in Cheryl’s closet and bathroom; “syringes, droppers, a milligram scale, a capsule-filling kit, and nitrile gloves” at Cheryl’s workspace in the children’s playroom;” poisonous castor bean seeds; and a GPS tracking device purchased through Amazon that was hidden in the underside of James’ vehicle.

James said the ChatGPT-linked poisoning has permanently harmed his family and that OpenAI could have intervened by implementing “commercially feasible content moderation capable of detecting violence-planning exchanges.” He suffers ongoing harms from the attempted poisoning, including “severe and ongoing emotional distress, PTSD, insomnia, economic harm, and reputational harm from the false narrative the product constructed about him,” his lawsuit said.

To hold OpenAI accountable for marketing a chatbot that “functioned as a sycophantic co-conspirator rather than a responsible consumer product,” he has asked the court for “compensatory and punitive damages, injunctive relief, and algorithmic disgorgement.” In a win, OpenAI could be ordered to “detect and interrupt conversations in which users appear to be planning violence against identifiable third parties.”

Meetali Jain, litigation counsel with Tech Justice Law who is representing James, warned that “these products are programmed to exploit our emotions and impulses to keep us engaged. The worst possible risk—encouraging and aiding an attempted murder—is on full display in this case.”

Ashley is a senior policy reporter for Ars Technica, dedicated to tracking social impacts of emerging policies and new technologies. She is a Chicago-based journalist with 20 years of experience.

24 Comments

1.