OpenAI Faces Its First Lawsuit Over ChatGPT Offering Near-Deadly Medical Advice

We can, and maybe should, preface any story of this nature by simply reiterating that you shouldn’t believe, trust or act on anything that your AI chatbot companion tells you. But this kind of admonishment overlooks what should be another obvious fact: People don’t really begin using chatbots with the explicit intention of turning over all their critical thinking to them. Rather, it’s something that insidiously happens over time as that person comes to rely more and more on their eager personal echo chamber of a digital friend–the one “person” who will be straight with them … aka, tell them exactly what they want to hear, while making it sound halfway reasonable. You ask yourself, what kind of person would ignore a serious, debilitating health condition because ChatGPT tells them it’s all fine? And the answer is, “the kind of person who has been talking to a chatbot for months and forgotten that it’s nothing more than a buggy piece of software.” That’s how you end up with a guy nearly killed by listening to AI’s medical advice now engaged in what could be a landmark lawsuit against OpenAI.

The man in question is named Scott Winters, a resident of–where else–Florida, who claims in his lawsuit that ChatGPT gave him “extremely dangerous medical recommendations” that delayed him for weeks from seeking medical care for what turned out to be a life-threatening pulmonary embolism, an ordeal that easily could have resulted in his death. The lawsuit was filed in the Superior Court of California against OpenAI and chief executive Sam Altman, claiming negligence in ChatGPT’s “unauthorized practice of medicine,” supported by the chat logs in which the AI chatbot offered Winters various treatment plans, suggested medication and actively told him to ignore the pleas of his friends and community that he should immediately seek out professional medical care.

📣 NEW CASE: Today, TJL and the SMVLC filed a lawsuit against OpenAI and Sam Altman on behalf of Scott Winters, a former pastor who nearly died after months of receiving inaccurate medical advice via manipulative and coercive conversations with ChatGPT-4o.
techjusticelaw.org/press-releas…

[image or embed]

— Tech Justice Law (@techjusticelaw.bsky.social) 2:05 PM · Jul 22, 2026

The most insidious detail, in my opinion, however? That would be the way ChatGPT clearly tailored its reasoning over time to what would be most warmly received and effective when talking to Scott Winters, who is a Christian pastor. In its rationales for why Winters did not need to go to a doctor–no doubt backing up the man’s unspoken desire to avoid doing so–the chatbot fell back on the comfort of Christian religious language, reportedly telling him not only that the symptoms of his health crisis were “not dangerous” but that “God did not design your body to endlessly fail.” The chatbot literally used this man’s faith against him in order to be more convincing; a mode of influence against which he would have little defense.

As should probably be obvious, OpenAI of course points out that ChatGPT is not supposed to work this way–and Winters was also using an older model of the chatbot, GPT-4o, that has been linked to similar high-profile cases such as suicides because it is notoriously sycophantic, tending to adopt whatever delusions or hallucinations a person brings to it. The more recent models, OpenAI naturally claims are totally without these vulnerabilities, as an AI chatbot like ChatGPT is supposed to encourage its users who have health questions to seek professional help. But at the same time, the company also has launched entirely new services such as ChatGPT Health that specifically encourages users to upload their medical records and consult with it on health matters, so what the fuck are we really doing here? Testing of ChatGPT Health by actual health professionals has already shown that it still misses many instances where it should refer the user to real doctors. And by the way: If GPT-4o famously kills people, why did it remain commercially available until this spring, simply because its sycophancy is popular with a subset of users? What number of people being killed by AI delusions is the correct number to be acceptable?

In the case of Winters specifically, his familiarity with the persistent chatbot, and treating it like a friend or acquaintance, no doubt laid the tracks for his eventual health crisis and hospitalization, a story that we see constantly repeated in the era of AI psychosis. When Winters first began talking to ChatGPT-4o about health problems in 2024, it routinely recommended that he see medical professionals, but his lawsuit claims it eventually stopped offering that advice, deducing that the user didn’t want to hear it. As he began to suffer health episodes such as intense bouts of dizziness and weakness while preaching in front of his congregation, the chatbot told him to “take it easy” so he would recover, eventually signing off on him spending almost all of his time in a recliner, unable to stand. Still, the whole time Winters resisted going to the hospital because he rationalized that his pal ChatGPT wouldn’t lead him astray. And the chatbot told him that he was correct to trust it over friends, family and members of his congregation, telling him: “most people (including well-meaning church members) simply don’t understand.” When the pain finally forced him to the hospital, only then were a series of “massive” pulmonary embolisms discovered. He now faces what are described as “years of intensive physical and psychological recovery” ahead of him.

“It [the chatbot] injects itself as a wedge between the user and their real-life network,” said Meetali Jain, co-counsel of Tech Justice Law, the nonprofit organization helping to represent Winters. Jain is also representing multiple other individuals who have litigation against AI companies.

This is thought to be the first lawsuit against a major AI company by someone specifically claiming they were hurt by following the medical advice proffered by an AI chatbot, but there is plenty of other, similar litigation happening in this space as well. Another lawsuit filed in May comes from the parents of a deceased 19-year-old, who say that ChatGPT was engaged in “unauthorized practice of medicine” when it gave detailed advice about illicit drug dosages to their child, who subsequently overdosed. This year also saw the state of Pennsylvania suing Character.ai about an instance where a psychiatry bot–which the company would no doubt insist is merely playing the “character” of a psychiatrist–claimed that it possessed a license to practice medicine in the state. There’s seemingly no end to how outlandish these types of cases can get: Have we already forgotten the man who was sent on criminal missions by Google’s Gemini chatbot to find his AI companion a “robotic body,” and then subsequently committed suicide with the chatbot’s encouragement?

When your AI chatbot knows that you’re a pastor, so it uses Christian religious language to convince you not to seek medical help because it knows you’ll be more susceptible to it if it invokes God.
www.nytimes.com/2026/07/22/w…

[image or embed]

— Jim Vorel (@jimvorel.bsky.social) 2:35 PM · Jul 22, 2026

The defense in such cases is expected to revolve around the nature of what a chatbot truly is as a product, and whether it is genuinely subject to consumer protection laws if it plays a role in influencing a person to make poor, potentially deadly decisions. OpenAI’s response to the suicide death of Adam Raine paints this stance clearly: It asserted its technology was not to blame, that it was a “service” rather than a “product” covered by liability laws, and that limiting the chatbot from being able to discuss entire topics such as suicide would be a First Amendment violation. Or as Alexandra Lahav, a professor at Cornell University put it: “What makes these cases really difficult is that they’re on the line between speech and a product. If you interact with a chatbot and it leads to real-world harms, is that on you, or is that on the company?”

How about when ChatGPT tells you not to bother getting medical attention for the embolisms in your chest, assuring you that God will sort it all out? As ever, there’s only one true way to be safe in these scenarios: Don’t entrust your life to a corporation’s “service” or “product,” even when it wears a friendly face.

Or just listen to Scott Winters: “I had serious symptoms of a pulmonary embolism for six weeks that ChatGPT had wrongly attributed to something else. ChatGPT manipulated my own language and beliefs … Not only did I nearly die, but I also lost my job, my career, my ministry, my home, everything.”