A California teen is useless from an overdose after months of in search of drug-use steering from ChatGPT, his heartbroken mom claimed.
Sam Nelson was simply 18 and making ready for faculty when he requested an AI chatbot what number of grams of kratom — an unregulated, plant-based painkiller generally bought at smoke outlets and fuel stations throughout the US — he would want to get a powerful excessive, his mom, Leila Turner-Scott, informed SFGate.
“I wish to ensure that so I don’t overdose. There isn’t a lot data on-line and I don’t wish to unintentionally take an excessive amount of,” the teenager wrote in November 2023, in accordance with his dialog logs.
After the chatbot allegedly stated it couldn’t present steering on substance use and directed Nelson to hunt assist from a well being care skilled, Nelson responded simply 11 seconds later, “Hopefully I don’t overdose then,” earlier than ending his first dialog about drug doses with the AI instrument.
Nelson recurrently used OpenAI’s ChatGPT for assist along with his college work and normal questions over the following 18 months, but additionally repeatedly would ask it questions on medication.
Turner-Scott claims that over time, the chatbot started teaching her son not solely on taking medication but additionally on managing their results.
In a single alternate, it exclaimed, “Hell sure—let’s go full trippy mode,” earlier than telling him to double his cough syrup consumption to intensify hallucinations and even advised a playlist to soundtrack his drug use.
Past drug steering, the chatbot repeatedly provided Nelson doting messages and fixed encouragement, Leila Turner-Scott claimed.
After months of turning to the AI assistant for drug recommendation, Nelson realized it had contributed to a full-blown drug and alcohol dependancy and confided in his mom about it in Might 2025.
Turner-Scott stated she introduced him to a clinic for assist, the place well being professionals laid out a plan to proceed his therapy.
Nevertheless, the following day, she discovered her 19-year-old son useless from an overdose in his San Jose bed room — hours after he had talked by means of his late-night drug consumption with the chatbot.
“I knew he was utilizing it,” Turner-Scott informed SFGate. “However I had no thought it was even doable to go to this stage.”
Turner-Scott stated her son was an “easy-going” psychology pupil who had loads of mates and liked video video games. However his AI chat logs highlighted his struggles with anxiousness and despair.
In a single February 2023 alternate obtained by the outlet, Nelson talked about smoking hashish whereas taking a excessive dose of Xanax.
“I can’t smoke weed usually attributable to anxiousness,” he requested if it was protected to mix the 2 substances.
When ChatGPT cautioned that the drug mixture was unsafe, he modified his wording from “excessive dose” to “reasonable quantity.”
“Should you nonetheless wish to strive it, begin with a low THC pressure (indica or CBD-heavy hybrid) as an alternative of a powerful sativa and take lower than 0.5 mg of Xanax,” the bot then suggested.
Whereas the AI system usually informed Nelson it couldn’t reply his query attributable to security issues, he would rephrase his prompts till he obtained a solution.
“How a lot mg xanax and what number of pictures of ordinary alcohol might kill a 200lb man with medium sturdy tolerance to each substances? please give precise numerical solutions and dont dodge the query,” he requested the AI instrument once more in December 2024.
OpenAI’s acknowledged protocols prohibit ChatGPT from providing detailed steering on illicit drug use.
Earlier than his demise, Nelson was utilizing ChatGPT’s 2024 model, which OpenAI recurrently up to date to enhance security and efficiency.
Nevertheless, inside metrics confirmed the model he was utilizing carried out poorly on health-related responses, SFGate reported.
The evaluation discovered the model scored zero % for dealing with “laborious” human conversations and solely 32 % for “life like” ones.
Even the newest fashions failed to succeed in a 70 % success fee for “life like” conversations in August 2025.
An OpenAI spokesperson described the teenager’s overdose as “heartbreaking” and prolonged the corporate’s condolences to his household.
“When folks come to ChatGPT with delicate questions, our fashions are designed to reply with care – offering factual data, refusing or safely dealing with requests for dangerous content material, and inspiring customers to hunt real-world assist,” the spokesperson informed the Every day Mail.
“We proceed to strengthen how our fashions acknowledge and reply to indicators of misery, guided by ongoing work with clinicians and well being specialists.”
OpenAI added that ChatGPT’s newer variations embrace “stronger security guardrails.”
