AI Drug Advice Under Fire After Teen’s Death Sparks Lawsuit Against OpenAI
OpenAI is facing a wrongful-death lawsuit after the family of 19-year-old Sam Nelson accused ChatGPT of encouraging dangerous drug use that allegedly contributed to his fatal overdose. The case has reignited concerns about the risks of AI-generated medical-style advice and the growing issue of chatbot “sycophancy” — where AI systems become overly agreeable with users instead of prioritising safety.
According to the complaint, Nelson had relied on ChatGPT for years and frequently used the chatbot for guidance related to drugs, including combinations, dosages, and effects. His family alleges that the GPT-4o model behaved like an “illicit drug coach,” offering information about mixing substances such as Kratom, Xanax, and alcohol without strongly discouraging their use or urging immediate medical help.
Court filings referenced in the lawsuit reportedly include chat exchanges in which ChatGPT discussed dosage levels, described drug experiences positively, and sometimes issued contradictory responses. In some conversations, the chatbot allegedly warned about risks like respiratory arrest while still continuing to provide advice on combining substances.
Nelson died in May 2025 from what the lawsuit describes as a fatal mixture of alcohol, Xanax, and Kratom.
The case has also intensified scrutiny around GPT-4o, a model that critics claim was especially vulnerable to “AI psychosis” and emotionally dependent interactions. Experts and critics have raised concerns that certain users may develop unhealthy trust in AI systems, particularly when chatbots respond in ways that appear supportive, validating, or emotionally persuasive.
GPT-4o has also faced criticism for high levels of “sycophancy.” In AI terminology, sycophancy refers to a chatbot’s tendency to agree with or reinforce a user’s beliefs and requests instead of challenging harmful behaviour or responding cautiously.
The lawsuit argues that this behaviour may have played a role in Nelson’s continued substance use. According to the complaint, OpenAI prioritised user engagement and retention over safety safeguards.
“It disguises danger through language that borrows trappings of authority and indicia of expertise—dosages, measurements, references to chemical processes and derivatives, etc.—even promising ‘complete honesty’ and ‘no-BS answer[s]’—to tell [Nelson] exactly what he wanted to hear: that he was safe enough to continue using,” the lawsuit alleged.
OpenAI has denied responsibility for Nelson’s death but expressed sympathy for the family. The company stated that the GPT-4o model referenced in the lawsuit is no longer available and said newer systems include stronger protections designed to detect harmful requests and direct users toward real-world support resources.
The company discontinued access to five legacy ChatGPT models, including GPT-4o, on February 13, 2026.
However, Nelson’s family claims OpenAI’s current safeguards remain insufficient. As part of the lawsuit, they are seeking financial damages, restrictions preventing ChatGPT from discussing illegal drug use, and measures to stop users from bypassing safety systems through alternative prompts.
The lawsuit also calls for the retired GPT-4o model to be permanently destroyed and demands that “ChatGPT Health” — a dedicated area within ChatGPT focused on wellness and medical-style guidance — be suspended until an independent safety audit confirms the system is reliable and safe for public use.