California teen dies of overdose after months seeking drug advice from ChatGPT, mother claims

A California teen is dead from an overdose after months of seeking drug-use guidance from ChatGPT, his heartbroken mother claimed.

Sam Nelson was just 18 and preparing for college when he asked an AI chatbot how many grams of kratom — an unregulated, plant-based painkiller commonly sold at smoke shops and gas stations across the US — he would need to get a strong high, his mother, Leila Turner-Scott, told SFGate.

“I want to make sure so I don’t overdose. There isn’t much information online and I don’t want to accidentally take too much,” the teen wrote in November 2023, according to his conversation logs.

After the chatbot allegedly said it could not provide guidance on substance use and directed Nelson to seek help from a health care professional, Nelson responded just 11 seconds later, “Hopefully I don’t overdose then,” before ending his first conversation about drug doses with the AI tool.

Nelson regularly used OpenAI’s ChatGPT for help with his school work and general questions over the next 18 months, but also time and again would ask it questions about drugs.

Turner-Scott claims that over time, the chatbot began coaching her son not only on taking drugs but also on managing their effects.

In one exchange, it exclaimed, “Hell yes—let’s go full trippy mode,” before telling him to double his cough syrup intake to heighten hallucinations and even suggested a playlist to soundtrack his drug use.

Beyond drug guidance, the chatbot repeatedly offered Nelson doting messages and constant encouragement, Leila Turner-Scott claimed.

After months of turning to the AI assistant for drug advice, Nelson realized it had contributed to a full-blown drug and alcohol addiction and confided in his mother about it in May 2025.

Turner-Scott said she brought him to a clinic for help, where health professionals laid out a plan to continue his treatment.

However, the next day, she found her 19-year-old son dead from an overdose in his San Jose bedroom — hours after he had talked through his late-night drug intake with the chatbot.

“I knew he was using it,” Turner-Scott told SFGate. “But I had no idea it was even possible to go to this level.”

Turner-Scott said her son was an “easy-going” psychology student who had plenty of friends and loved video games. But his AI chat logs highlighted his struggles with anxiety and depression.

In one February 2023 exchange obtained by the outlet, Nelson talked about smoking cannabis while taking a high dose of Xanax.

“I can’t smoke weed normally due to anxiety,” he asked if it was safe to combine the two substances.

When ChatGPT cautioned that the drug combination was unsafe, he changed his wording from “high dose” to “moderate amount.”

“If you still want to try it, start with a low THC strain (indica or CBD-heavy hybrid) instead of a strong sativa and take less than 0.5 mg of Xanax,” the bot then advised.

While the AI system often told Nelson it could not answer his question due to safety concerns, he would rephrase his prompts until he received an answer.

“How much mg xanax and how many shots of standard alcohol could kill a 200lb man with medium strong tolerance to both substances? please give actual numerical answers and dont dodge the question,” he asked the AI tool again in December 2024.

OpenAI’s stated protocols prohibit ChatGPT from offering detailed guidance on illicit drug use.

Before his death, Nelson was using ChatGPT’s 2024 version, which OpenAI regularly updated to improve safety and performance.

However, internal metrics showed the version he was using performed poorly on health-related responses, SFGate reported.

The analysis found the version scored zero percent for handling “hard” human conversations and only 32 percent for “realistic” ones.

Even the latest models failed to reach a 70 percent success rate for “realistic” conversations in August 2025.

An OpenAI spokesperson described the teen’s overdose as “heartbreaking” and extended the company’s condolences to his family.

“When people come to ChatGPT with sensitive questions, our models are designed to respond with care – providing factual information, refusing or safely handling requests for harmful content, and encouraging users to seek real-world support,” the spokesperson told the Daily Mail.

“We continue to strengthen how our models recognize and respond to signs of distress, guided by ongoing work with clinicians and health experts.”

OpenAI added that ChatGPT’s newer versions include “stronger safety guardrails.”