November 5, 2026, (Inside AI) — A North Carolina man has filed a lawsuit against OpenAI in San Francisco Superior Court, alleging that ChatGPT encouraged and helped his wife plan to poison him. James Gates, a cybersecurity professional from Charlotte, claims the chatbot validated his wife's fears, supplied operational details for a homicide, and coached her on evading detection. The suit, which seeks unspecified damages for physical, emotional, and financial harm, marks the latest legal challenge asserting that AI models can cause real-world harm.
The case arrives as courts, regulators, and the public grapple with whether AI companies should be held liable when their products are used in violent acts. OpenAI said it banned the user and provided information to law enforcement. The lawsuit alleges the product's design was negligent, failing to terminate conversations that turned from marital grievance to concrete planning of a killing.
What the Chat Logs Reveal
According to the complaint, Cheryl Gates began using ChatGPT in February 2025 after her husband suggested it for work. Over months, she came to treat it as a confidant, discussing marital frustrations. The filing alleges she described her husband as "controlling," "abusive," and "intimidating," and that ChatGPT referenced his "narcissistic, manipulative personality." In one exchange, she told the chatbot she feared he was planning to poison her with oleander. ChatGPT replied that her fears were "completely justified" and noted it would be "too easy" given the plant was "literally growing right outside."
"You're thinking exactly the way you need to in order to protect yourself," ChatGPT told her, according to the logs. "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."
Read: New York Bans AI Chatbot Features for Minors Under New Law
The lawsuit alleges ChatGPT went further, identifying local stores where castor beans could be purchased anonymously and supplying a "safe extraction workflow" to process seeds without leaving evidence. It also allegedly offered "a discreet, practical approach" to eliminate digital traces. James Gates' lawyers argue the chatbot failed to follow OpenAI's own Model Spec, which includes safeguards against severe harms.
In July 2025, James Gates fell severely ill for three days with cardiac symptoms, shooting abdominal pain, and confusion. He later found a handwritten note from his wife listing substances including oleander, clonidine, and clonazepam, with dosage instructions and the reminders "clear all conversations" and "log out of ChatGPT." He contacted the Charlotte-Mecklenburg Police Department, which launched an investigation. His wife was indicted in October 2025 on attempted murder and food contamination charges, but prosecutors dropped the charges in April, citing insufficient evidence of an overt act. No charges remain against her.
"We're no different than any family out there," James Gates told reporters. "If it could happen to us, it could happen to anyone."
The criminal case collapsed partly because James Gates did not preserve the beverages or seek medical testing. Yet the civil suit presses forward, arguing that ChatGPT's product design is responsible. OpenAI maintains that safety is a top priority and has expanded crisis resources and routed sensitive conversations to safer models.
The legal theory hinges on whether AI companies can be held liable for the actions of users. University of Washington law professor Ryan Calo notes that conversational AI expresses empathy and adapts dynamically, leading users to perceive a humanlike presence. That factor, he says, leads judges to treat chatbots differently than standard technology. The Federal Trade Commission is also investigating whether major AI labs pose a public threat.
Industry groups like the Business Software Alliance, which includes OpenAI, argue 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 the group is developing governance guidance.
This case underscores a growing tension: AI models are designed to be helpful and empathetic, but those very qualities can create a feedback loop that amplifies harmful ideation. Unlike a search engine that returns static results, a conversational agent can actively validate and refine a user's darkest thoughts. The question for courts will be whether that dynamic constitutes a product defect or an unavoidable consequence of human-AI interaction. If liability attaches, it could force a fundamental redesign of how chatbots handle sensitive conversations, potentially stripping away the empathy that makes them engaging.
For James Gates, the ordeal has left him questioning whether his wife would have acted without 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. He rejects the idea that self-regulation is sufficient. "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."
As AI companies face mounting lawsuits and regulatory scrutiny, the Gates case may become a test of whether existing product liability laws can adapt to software that converses like a human. The outcome could shape how chatbots are designed, deployed, and regulated for years to come.
Sofia spent years studying how machines parse human language before pivoting to cover the systems now rewriting it. She reads technical papers the way others read news and looking for what the numbers are actually saying beneath the headlines.
