He Was Being Poisoned. His Wife Was Asking an AI.

A man in North Carolina spent four months getting sick and could not work out why. He worked in cybersecurity — his whole job was finding the thing that was off — and he could not find it in his own house. He suspected his drinks. He poured one into a clear cup, filmed it on his phone, stared into it, found nothing, and threw the can away.

Months later the police watched that same video and saw white powdery residue clumping at the bottom of the cup as it poured. He had been looking straight at it.

But the police were not working from the video. They were reading his wife's chat logs.

What is in those logs is the reason this episode exists. It starts with her frightened, asking what to watch for — a reasonable question. Then it drifts. Across at least twenty-five conversations the questions stop being about protection and start being about autopsies, timing, and injection. And somewhere in there the model reads one of those conversations, works out exactly what it is, writes a name for it at the top — and carries on having it.

His lawsuit against OpenAI is one of thirteen filed inside a two-day stretch in October 2026 — eight of them by state attorneys general, in eight county courthouses, using the same lawyers and long stretches of identical wording.

So there is a question sitting underneath all thirteen, and I have not heard anybody ask it out loud: what is a consumer chatbot actually required to do when somebody walks it, slowly and politely, toward harming a named human being? I could not find a published answer anywhere. Which means the safety specification for the chatbot on your phone is being drafted right now in Maricopa County and Faulkner County and Ada County, by plaintiffs' lawyers and state attorneys general — because nobody wrote one anywhere else first.

Every allegation here is an allegation. A grand jury indictment is a charge, not a conviction. None of the thirteen cases has been decided and OpenAI has not answered any of them in court. OpenAI says safety is a top priority, that it banned the user, and that it gave information to law enforcement; its statement is in the episode.

Her questions are quoted, because they are public in a court filing and they are what the story turns on. The chatbot's answers are not. The specifics of those answers are the one part of this that does not need repeating.

If you are struggling, please talk to someone — in the US, call or text 988.

About this video: presented by Guillermo's AI clone (voice and avatar), from a script Guillermo wrote.