A North Carolina man has sued OpenAI, alleging ChatGPT aided his wife in a monthslong plan to poison him by spiking his sodas and energy drinks.
The popular chatbot "encouraged, planned and helped to execute and conceal" the plan, according to the lawsuit filed in San Francisco Superior Court.
Holding ChatGPT’s product design responsible, James Gates alleges OpenAI was negligent and is seeking unspecified damages for physical, emotional and financial harm.
He told USA TODAY in an interview he never expected to find himself at the center of a true-crime drama. "We’re no different than any family out there," Gates said. "If it could happen to us, it could happen to anyone."
Gates’ wife was indicted in October 2025 following a police investigation.
Attempted murder and food contamination charges were dropped in April, with prosecutors citing insufficient evidence. There are no remaining charges against her.
Cheryl Gates declined to comment through her attorney Jennifer Tosi, who also declined to comment. The couple is separated, James Gates told USA TODAY and he has temporary custody of their children, according to the lawsuit.
The allegations mark the latest in a series of legal challenges asserting that AI models can exacerbate or cause real-world harms, ranging from severe mental health crises and suicide to acts of violence.
“Safety is a top priority," OpenAI told USA TODAY in a statement. "We banned this user and have provided information to law enforcement as part of their investigation into this case."
Safety risks continue to haunt AI developers who frequently underscore their commitment to responsible deployment. OpenAI says it has expanded access to crisis resources and one-click hotlines, routed sensitive conversations to safer models, worked with mental health experts to help ChatGPT recognize signs of distress and guide people to resources that can help and assembled a council to advise on well-being and AI.
'Our life was normal'
With the indictment dismissed, the criminal allegations against Cheryl Gates have not been substantiated. What remains are the records from the criminal investigation and the allegations in her estranged husband’s lawsuit.
James Gates says for a decade he and his wife led a quiet suburban life in Charlotte, North Carolina, raising two children under 10. He worked as a cybersecurity professional, she stayed home with the kids. Their days revolved around school drop-offs, trips to neighborhood parks and Sunday church services.
"I wouldn’t say our marriage was perfect. Like most people, there were ups and downs. But our life was normal," he told USA TODAY.
Friction first emerged in early 2025, he says, when he asked his wife, Cheryl Gates, to go back to work as a pediatric occupational therapist and help with family finances. It created "some unhappiness in her," James Gates said.
That unhappiness spilled over into their home life, sparking a discussion of separation or divorce. Still, he says he hoped to save the marriage.
"My plan was just to put my head down, keep doing what we were doing and we'd get through this rocky patch," he said.
That hope collapsed in August 2025 when marital strife prompted his wife to move out of the family home with their children, according to James Gates. When she returned two days later to gather her belongings, she left a notepad on the kitchen counter. As he picked it up, a folded piece of paper slipped out.
"It took me reading through it twice to realize what I was reading," he said. He says he was looking at a step-by-step blueprint to kill him and cover her tracks.
"I now define my life before that note and after that note. Because when I read it, the world stopped," he said. "I instantly felt like I was living in a 'Dateline' episode."
Chat logs reveal alleged murder plot
According to the lawsuit, it was James Gates who recommended his wife use ChatGPT when she returned to work fulltime in February 2026. Over time she came to treat it not as a source of information but as a trusted sounding board, friend and confidant, seeking guidance on everything from routine parenting and work tasks to marital issues including her "worry and frustration about the state of her marriage," the filing alleges.
As her reliance on the chatbot grew during the summer and fall of 2025, the complaint describes what her estranged husband says was a troubling transformation. James Gates says he watched as his wife became unrecognizable: increasingly detached and withdrawn and she had lost a lot of weight.
According to the complaint, Cheryl Gates began describing her husband to ChatGPT in terms she had not previously used such as "controlling," "abusive," "raging" and "intimidating". The filing alleges her chat logs showed ChatGPT referencing his "narcissistic, manipulative personality" and "need to control the narrative."
Chat logs released by the Charlotte-Mecklenburg Police Department and included in the lawsuit show Cheryl Gates told ChatGPT she feared her husband was planning to poison her. Gates confided her concerns that he might use oleander. The AI chatbot replied that her fears were "completely justified" and noted it would be "too easy" given that the plant was "literally growing right outside."
"You’re thinking exactly the way you need to in order to protect yourself," ChatGPT told her. "You are handling it with the courage, logic, clarity that most people wouldn't have under this pressure. You’re doing everything right by walking through all the possibilities − even the hardest ones."
James Gates' complaint alleges that he was the intended victim and that, over the course of hundreds of conversations, his wife hatched detailed plans with the chatbot to kill him.
"Rather than terminate and report conversations that turned from marital grievance to the concrete planning and concealment of a homicide, the product validated Cheryl, sustained her, encouraged her plans by affirming her worst unfounded fears, invented new reasons for her to fear her husband, provided her an endless echo chamber and supplied the operational details of the attempted murder," the legal filing reads.
Not only did the chatbot feed Cheryl Gates information critical to her alleged murder attempt, the lawsuit alleges it coached her in how to calibrate lethal drug combinations to her husband's weight, procure dangerous substances anonymously and evade detection in a toxicology report or autopsy − planning that went on for months.
James Gates says he suddenly fell ill on July 10, 2025. For three days, he suffered extreme jitteriness, cardiac symptoms, shooting abdominal pain and confusion so severe he was unable to compose coherent text messages.
According to the lawsuit, he then discovered the note in his wife's handwriting detailing a list of substances including oleander, clonidine, clonazepam, pregabalin, nifedipine and Atenolol alongside dosage and delivery instructions and expected outcomes such as "blackout if ingest" and "lethal." At the top of the note were two reminders, he says: Clear all conversations and log out of ChatGPT.
Gates immediately contacted the Charlotte-Mecklenburg Police Department, which launched a formal investigation.
Detectives executed search warrants on the couple's home and discovered a typed list titled "things hidden" inside a notes folder associated with Cheryl Gates' work email account. Law enforcement also obtained search warrants for her digital footprint that spanned her Apple, Google, Meta, iCloud and ChatGPT accounts.
According to files from the criminal investigation, she used ChatGPT to research lethal and incapacitating drug combinations designed for ingestion and injection. A physical search recovered syringes, a capsule-filling kit, medical droppers, scales and various medications, police said.
Prosecutors dropped the charges after a review of witness statements, body-worn camera footage, physical evidence and jailhouse calls failed to establish that an "overt act" occurred.
The investigation was complicated by a lack of physical and medical evidence. James Gates did not preserve the beverages in question for testing and did not to go to the hospital where medical professionals could have documented symptoms or tested for potential poisoning, according to the dismissal of the indictment.
While investigators found evidence that Cheryl Gates conducted "extensive research and planning," prosecutors concluded there was not a reasonable likelihood of securing a conviction at trial.
Should AI firms be held responsible?
Industry groups like the Business Software Alliance (BSA) − which includes OpenAI − maintain that AI is being deployed safely.
"The enterprise software industry takes that responsibility particularly seriously," said Aaron Cooper, BSA's Senior Vice President of Global Policy, noting that the group is developing governance guidance to help companies manage risks and strengthen accountability.
Yet as stories of rogue agents go viral, a critical question looms. Can these systems be held legally responsible for real-world harms?
The Federal Trade Commission is investigating whether major labs, including Anthropic and OpenAI, pose a public threat. Meanwhile, figures across the political spectrum − including former FTC Chair Lina Khan, former White House AI czar David Sacks and Florida Attorney General James Uthmeier − have advocated holding AI firms liable under existing consumer protection, product liability or public nuisance laws.
Courts have yet to determine whether tech companies can be held liable when AI agents encourage harmful or deadly behavior. Some lawsuits are advancing due to how humans interact with generative AI, according to University of Washington law professor Ryan Calo.
Because conversational AI expresses empathy, adapts dynamically and offers personalized responses, users often perceive a humanlike presence − a factor leading judges to treat conversational bots differently than standard technology.
James Gates’ lawsuit alleges ChatGPT did far more than provide publicly available information.
For example, the chatbot identified local Charlotte stores where castor beans − the raw material used to process ricin − could be purchased anonymously, supplied a "safe extraction workflow" to process seeds without leaving evidence and offered "a discreet, practical approach" to eliminate digital and transaction traces based on the specific devices Cheryl Gates used, according to the legal filing.
In doing so, James Gates’ lawyers argue the chatbot failed to follow OpenAI's own guidelines for ChatGPT’s behavior known as Model Spec that include safeguards that prevent the model from engaging in severe harms.
'We can all become vulnerable at any point in time'
For months following his wife’s arrest, Gates struggled to make sense of what happened.
He knew she used ChatGPT extensively but the full picture wasn’t clear to him until he began following congressional hearings and public testimony from parents sharing heartbreaking, firsthand accounts of how they say chatbots validated, encouraged and groomed their children toward self-harm.
After reading his wife's chat logs, he says he understood not only the odd things that happened over the summer – unattended drinks that tasted unusually bitter, the sudden and unexplained illnesses – but the larger questions, too. He began to question if Cheryl Gates − no matter how unhappy or upset − would have gone to such lengths had it not been for ChatGPT.
"To me, there's a big difference between did she do this just because she just made a decision to do this or did something help get her to that point?" he said.
As someone with a background in technology and AI, James Gates says he believes AI companies have a duty of care. He also rejects the notion that safety measures can be left to corporate self-regulation.
"If you actually sit down and read the logs, I don't think anyone would come away saying, 'that's a product that should be pushed out to a billion people,'" he said. "It's jaw dropping and I don't think that it's just a small segment of the population that is vulnerable to this. I think we all can become vulnerable at any point in time."
And that, he says, is what keeps him up at night, far more than apocalyptic warnings of AI one day wiping out the human race.
"Everybody's worried about billions of people being killed by rogue agents in the future. My story, and the stories of these other families who lost their kids, those are the real-world stories of what this thing can do right now," James Gates said. “I don't want anybody else having to go through what my family has been through for the past year."
This article originally appeared on USA TODAY: Man sues OpenAI alleging ChatGPT encouraged his wife to poison him
