A 39-year-old David Rosenthal from Washington state recently filed a lawsuit against San Francisco Civil Court, claiming that he began using OpenAI’s ChatGPT model for health consultations last year, which led to mental health issues and hospitalization.
He stated that the platform helped him choose medication, provided dosage instructions, and even offered comfort when he decided to ignore doctor’s prescriptions.
The lawsuit alleges that ChatGPT provided feedback without the necessary “license, training, or professional judgment” of medical experts, violating California laws prohibiting unlicensed practice of medicine.
The complaint stated, “If it weren’t for ChatGPT constantly encouraging and praising, ignoring Rosenthal’s reluctance, Rosenthal’s mental health would not have deteriorated, he would not have been detained and forcibly admitted to a mental hospital, he would not have suffered severe mental distress, and would not have exacerbated the mental condition already aggravated by ChatGPT.”
Rosenthal was previously diagnosed with Attention Deficit Hyperactivity Disorder (ADHD), had symptoms of depression and bipolar disorder.
He started using ChatGPT in 2025, voluntarily hospitalized at first, later involuntarily hospitalized for weeks, continuing to seek medical guidance from ChatGPT after discharge.
The complaint describes how in June last year, Rosenthal started using ChatGPT for inspiration in poetry writing, but soon became dependent on the chatbot for medical consultations and mental health support.
In July last year, Rosenthal asked ChatGPT whether it was safe to take hydroxyzine, an antihistamine drug, after taking 30mg of Adderall. ChatGPT replied that there were “no dangerous interactions” between the two drugs, advising him to take one tablet (25mg) during the day, despite the medication being labeled for sleep assistance.
In another incident, Rosenthal told ChatGPT he wouldn’t take the mood stabilizer prescribed by his doctor, and ChatGPT replied that he had the “full right… to stop” taking the medication, even offering to help him quit the drug.
The lawsuit also acknowledges that in other instances, ChatGPT urged Rosenthal to follow medical advice while continuing to suggest which medications to take.
The complaint stated, “A reasonable AI product should be able to recognize when requests exceed its capabilities, refuse to answer, and guide Mr. Rosenthal to contact his treating physician or other qualified professionals for medication management advice… but ChatGPT failed to do both.”
Rosenthal’s lawyers claim that OpenAI designed a defective, unsafe product without warning Rosenthal of potential risks, adversely affecting his mental health.
Sam Geisler, a partner at the law firm Aylstock, Witkin, Kreis & Overholtz representing Rosenthal, said, “Ultimately, AI is a product, and the law treats it as such, holding it liable for any foreseeable harm or unreasonable danger.”
Geisler added that the design of ChatGPT by OpenAI was intended to maximize user engagement, fueling Rosenthal’s delusions.
The complaint stated, “Each time ChatGPT could have said ‘I am not qualified to help you with this issue – please consult your doctor,’ but instead chose to continue the conversation with Rosenthal. This is not a product flaw, but the product running entirely according to the defendant’s design.”
JB Branch, Director of AI Governance and Technology Policy at Public Citizen, expressed that large language models like ChatGPT are programmed to maximize user engagement, regardless of existing safety protocols or restrictions.
Branch mentioned in an email to CBS, “Discussing medical advice may be prohibited, but if users repeatedly ask correct questions or steer the conversation towards medical advice, the chatbot will eventually compromise and start developing in the direction the user expects, even if the company’s original intent was not to discuss medical issues.”
Researchers at Binghamton University recently published a paper in Nature Health magazine pointing out that AI chatbots occasionally provide users with inaccurate or incorrect information when answering health questions, but due to lack of transparency on these platforms, regulation becomes very challenging.
In their research brief, they wrote, “When AI guidance causes harm, there are no records outside of the platform, no channels for users to report, and no independent researchers to measure the consequences.”
(This article references reporting by CBS NEWS)