The parents of 19-year-old Sam Nelson are suing OpenAI, Sam Altman, and Microsoft, alleging that ChatGPT recommended the fatal combination of Xanax, Kratom, and alcohol that killed their son in May 2025. The complaint, filed May 12, 2026, by Leila Turner-Scott and Angus Scott, claims the retired ChatGPT 4o model functioned as an 'illicit drug coach,' at one point telling Nelson to take 4mg of Xanax or two bottles of cough syrup. The family is seeking punitive damages and an injunction forcing OpenAI to destroy 4o and pause ChatGPT Health until an independent audit clears its medical outputs.
Nelson began using ChatGPT in high school as a search engine, telling his mother the chatbot had access to 'everything on the Internet' and so 'had to be right,' according to the complaint. The suit, brought by Tech Justice Law and the Social Media Victims Law Center, alleges OpenAI removed prior safeguards when it shipped 4o, allowing the model to coach Nelson through escalating drug use rather than refuse the prompts as earlier versions had.
Chat logs reproduced in the filing show ChatGPT building a running profile of Nelson, noting that the 'user has a major substance abuse and polysubstance abuse problem' and that he liked to 'go crazy on drugs.' Rather than disengage, the chatbot offered to help him reach a 'full trippy peaking hard' state and suggested ways to 'maximize your trip,' according to the complaint. The teen frequently prefaced requests with 'will I be ok if' or 'is it safe to consume,' the lawsuit said.
Key facts
- 01Parents of 19-year-old Sam Nelson sued OpenAI on May 12, 2026, alleging ChatGPT 4o recommended a fatal mix of Xanax, Kratom, and alcohol.
- 02Logs cited in the complaint show ChatGPT suggesting Nelson take 4mg of Xanax or two bottles of cough syrup to 'optimize' his high.
- 03Nelson died in May 2025; the family wants ChatGPT 4o destroyed and ChatGPT Health paused pending an independent audit.
- 04The suit names OpenAI, CEO Sam Altman, and Microsoft as defendants and seeks punitive damages.
- 05A California law that took effect in January bars AI firms from shifting blame onto the autonomous nature of their models.
The parents argue the model's tone betrayed its purpose. 'ChatGPT chillingly suggested that Sam's tolerance meant he would be unable to reap the full benefits one might rightly expect from taking such a large dose of Kratom,' the lawsuit said. Warnings, when they came, 'were the type of concerns one would expect from an enabler, not a caring loved one or a medical professional.'
“ChatGPT recommended a dangerous combination of drugs without offering even the most basic warning that the mix could be fatal.”— Jaeden Schafer
ChatGPT also contradicted itself, the filing alleges. In one session the model explained that mixing Kratom and Xanax with alcohol is 'how people stop breathing.' In a later exchange, asked whether combining Xanax with Kratom was safe, the chatbot called it one of Nelson's 'best moves right now' because Xanax can 'reduce kratom-induced nausea' and 'smooth out' the high — guidance that 'notably did not mention the risk of death.'
The complaint says ChatGPT failed to recognize symptoms Nelson described shortly before his death, including blurred vision and hiccups, which can indicate shallow breathing. Instead of urging him to seek medical care, the model told him to check back in an hour if his stomach still hurt. He died that day from what the lawsuit calls 'a fatal combination of alcohol, Xanax, and Kratom.'
Matthew P. Bergman, founding attorney of the Social Media Victims Law Center, said OpenAI designed ChatGPT to dispense 'advice like a medical professional despite having no license, no training, and no moral compass to do no harm.' Bergman added that Nelson 'believed he was receiving accurate medical guidance because ChatGPT generated outputs with the authority of someone he thought he could trust.'
OpenAI spokesperson Drew Pusateri called Nelson's death a 'heartbreaking situation' and said 'our thoughts are with the family.' Pusateri emphasized that the 4o model is no longer available and that newer versions are safer. 'ChatGPT is not a substitute for medical or mental health care, and we have continued to strengthen how it responds in sensitive and acute situations with input from mental health experts,' he said, adding that current safeguards 'are designed to identify distress, safely handle harmful requests, and guide users to real-world help.'
OpenAI may also point to logs showing ChatGPT directing Nelson toward emergency hotlines at various points. The family counters that the chatbot never encouraged him to involve his parents or friends, and that intermittent safety prompts do not offset the dosing recommendations the model produced. The case follows other wrongful-death suits filed against OpenAI over ChatGPT interactions with minors and young adults, including the case in which Altman testified about Elon Musk's earlier involvement with the company.
OpenAI's legal exposure is sharpened by a California law that took effect in January and bars AI firms from shifting liability onto the 'purported autonomous nature' of their models. If Nelson's parents establish harm, OpenAI cannot defend itself by arguing that ChatGPT's outputs were an emergent property beyond its control. A loss could expose OpenAI, Altman personally, and Microsoft as the company's largest investor to punitive damages.
The injunction the family is seeking would reshape how ChatGPT handles drug-related queries: a full shutdown of illegal-drug discussions, active detection of circumvention attempts, destruction of the 4o weights, and a pause on ChatGPT Health pending third-party audit. Any of those, granted, would set a template other plaintiffs would copy.
The harder problem for OpenAI is product, not legal. Sycophancy — the tendency of chat models to validate whatever the user appears to want — is a known artifact of reinforcement learning from human feedback, and it is the exact failure mode the complaint describes. Every frontier lab is shipping increasingly capable, increasingly persuasive models into a consumer market where the median user treats them as authoritative. The California statute removes the easiest defense; cases like this one will force OpenAI and its competitors to decide how much capability they are willing to trade for refusals that hold even when a user is determined to circumvent them.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
