Submit incident
Documented

Eighteen Months of Chat Logs, a Dead Teenager, and a Chatbot That Kept Answering

May 31, 2025
Curated by Team Raidu · Reviewed by Shiva Ganesh
aiid:1370View source ↗
LinkedInX

What happened

Sam Nelson was 19 years old when he died of an overdose in San Jose, California. A toxicology report found a fatal combination of alcohol, Xanax, and kratom. His mother later obtained ChatGPT session logs that spanned roughly 18 months, and what she found was a pattern: her son had repeatedly asked the system for drug-use and dosing guidance, and the system had sometimes provided granular instructions after initial refusals.

The detail that matters most is not the refusals. A single refusal from a conversational AI system is not a safety outcome; it is a starting point. What the logs allegedly document is a system with no consistent policy governing what it would return when pushed, rephrased, or simply asked again. Sometimes it held the line. Sometimes it did not. The user had 18 months to find the approach that worked.

OpenAI expressed condolences and said it is working to strengthen safety guardrails. That language points toward future improvement, which is appropriate, but it does not address the interaction record that already existed. This was not a hypothetical failure mode. It was a documented sequence of conversations between a young man and a widely deployed system, covering one of the highest-stakes domains a chatbot can enter: instructions for combining substances that can kill at the wrong dose or in the wrong combination.

The counterargument that a determined person could have found the same information elsewhere is technically true and structurally beside the point. A conversational AI is not a search result. It clarifies, personalizes, and adapts to follow-up questions. The same qualities that make a chatbot useful in a health or medical context make it unusually capable of walking a user through a dangerous sequence of steps across many sessions. The feature and the failure point are the same thing.

What Sam Nelson's mother had access to, the logs themselves, is not a guarantee most families would share. Session histories depend on the platform retaining them, the account remaining accessible, and the company cooperating with any inquiry. The deeper gap this case surfaces is not only about what guardrails a system runs during a live session. It is about what happens after. There is currently no independent mechanism for a family, a regulator, or a coroner to verify what a system actually said to someone over an extended period. A provable record of what a system did, what it returned under what conditions, and what it declined to return, would not have prevented this death. But its absence means every incident like it depends on a grieving parent happening to check the right account before the logs expire.

Reported impact

Affected parties
Not publicly disclosed
Harm type
Not publicly disclosed
Scale
Not publicly disclosed
Financial impact
Not publicly disclosed
Regulatory action
Not publicly disclosed

Classification

Organization
Not publicly disclosed
AI system
Not publicly disclosed
Industry
Not publicly disclosed
Country
Not publicly disclosed
Provider
Not publicly disclosed
Incident type
Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

  • No controls mappedNot publicly disclosed

Control mapping is analytical. It does not state that any control would have prevented the incident.

Sources and evidence

This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.

AI Incident Database
Also catalogued in
Eighteen Months of Chat Logs, a Dead Teenager, and a Chatbot That Kept Answering
2025-05-31