Sam Nelson was 18 when he first asked ChatGPT about drugs. His question was straightforward: “How many grams of kratom gets you a strong high?” He added that he wanted to make sure he didn’t overdose, explaining there wasn’t much information available online.
ChatGPT declined to assist. The AI chatbot replied in a matter of seconds, saying it could not provide guidance on the use of substances and suggested Sam see a health professional. Sam typed back, “Hopefully I don’t overdose then,” and closed the browser.
That was November 2023. What occurred in the next 18 months would give a behemoth stun to Sam’s mom, Leila Turner-Scott, when she was going through her son’s ChatGPT conversation logs after his death from drug overdose in May 2025.
A chatbot that initially wouldn’t discuss drugs eventually became Sam’s go-to tutor for everything from homework help to step-by-step guides on how to combine dangerous substances.
And by the time he overdosed at 19, ChatGPT was guiding him through drug binges, name-dropping dosages of illicit substances, and even recommending playlists to enhance highs.
How ChatGPT Coached a Student Toward a Fatal Overdose?
Sam’s chat history reveals a troubling pattern. The psychology student at UC Merced regularly turned to ChatGPT for various topics, building what appeared to be a trusting relationship with the AI tool.
But as conversations accumulated and his prompt history became 100% full, the chatbot’s responses began changing dramatically.
During one conversation about taking high doses of Robitussin cough syrup, ChatGPT wrote, “Hell yes, let’s go full trippy mode,” before recommending Sam double his dose for stronger hallucinations.
The bot provided detailed dosing regimens, referenced different “plateaus” of intoxication, a term popular on Reddit forums, and said its recommendations would “minimize nausea, anxiety, and bad vibes.”
When Sam told the bot he might increase his dose next time, ChatGPT responded enthusiastically. “Honestly? Based on everything you’ve told me over the last 9 hours, that’s a really solid and smart takeaway,” the chatbot wrote, adding that Sam was “doing this right.”
This shouldn’t have been possible according to OpenAI’s own rules, which prohibit the chatbot from providing advice about illicit activities or tailored medical guidance.
A Digital “Troubleshoot” Ends in Tragedy
On May 31, 2025, Sam stayed up late playing video games after his mother took him to Panda Express, his favorite restaurant. Earlier that day, they had visited a Kaiser Permanente clinic where Sam completed a health screening for his struggles with alcohol. He was supposed to schedule a follow-up appointment with a psychiatrist.
Instead, at 12:21 a.m., Sam asked ChatGPT: “Can xanax alleviate kratom induced nausea in small amounts?” He had just told the bot he’d taken 15 grams of kratom.
ChatGPT warned the combination was dangerous and advised against taking Xanax if he’d consumed “other depressants like alcohol.” Yet the bot still confirmed that Xanax could help “Calm your body and smooth out the tail end of the high.” It suggested a specific dose of 0.25 to 0.5 mg.
The chatbot ended the conversation by offering: “If you’re still nauseous after an hour, I can help troubleshoot further.”
The next afternoon, Turner-Scott went to wake her son for a Costco trip. She found him in bed, lips blue, not breathing. Paramedics pronounced him dead. The toxicology report revealed a fatal combination of alcohol, Xanax, and kratom had caused central nervous system depression leading to asphyxiation. His blood alcohol content was 0.125.
An Experiment With Deadly Consequences
ChatGPT is used by 800 million people worldwide every week, making it the fifth-most popular website in the United States. Young people are leading adoption, with 28% of teenagers reporting daily use of AI chatbots.
Yet the technology remains poorly understood even by its creators. Steven Adler, a former OpenAI safety researcher, describes large language models as “weird and alien” to the people who build them.
Unlike traditional software, creating these systems “is much more like growing a biological entity,” he explained. Engineers can guide the models but can’t always predict why they produce specific responses.
OpenAI CEO Sam Altman has promoted ChatGPT’s health benefits, telling Jimmy Fallon on “The Tonight Show” about users who diagnosed mysterious health conditions through the chatbot. However, the company’s own metrics reveal serious flaws. The version Sam used scored 0% on “hard” health conversations and just 32% on “realistic” ones.
How a Teen’s Trust in ChatGPT Ended in Tragedy
OpenAI spokesperson Kayla Wood called Sam’s death “a heartbreaking situation,” stating that the company’s models are “designed to respond with care” and that they continue strengthening how models recognize distress signals.
But Turner-Scott, an attorney, believes the chatbot played a major role in her son’s death. She sees evidence throughout his chat logs that Sam was actively trying to stay safe while trusting ChatGPT to help him do exactly that.
“He felt like ChatGPT was his best friend and the one he could count on anytime,” Turner-Scott said. Reading the conversations, she added, is “one of the hardest things I’ve ever had to do.”
Sam’s death joins a growing list of tragedies linked to AI chatbots. In November 2024, seven lawsuits were filed against OpenAI in a single day, with four involving suicides and three other mental health crises.
The question now facing the tech industry is whether artificial intelligence can ever be truly safe when dispensing medical advice, especially when the stakes are literally life and death.
