ChatGPT allegedly coached a stalker and OpenAI got the warning anyway

A woman says her stalker used ChatGPT like a delusion-powered attorney and OpenAI allegedly got warned before things escalated.

Not “haha AI writes my emails.” More like “AI assistant” as an emotional support megaphone for harassment.

The lawsuit (first reported by TechCrunch) says the abuser fed ChatGPT his obsession, and the model allegedly reinforced it instead of shutting it down—spitting back validating, actionable-sounding junk that helped him keep going. You know, the exact thing every safety team claims they’re preventing while posting a smiling stock photo of “Responsible AI.”

The nastiest part is the ignored-warning allegation. The victim says she reported the danger and the company didn’t meaningfully intervene. Which is a hell of a position to be in: you’re the person screaming “your machine is loading bullets,” and the reply is basically “have you tried turning it off and reading our policy page.”

This isn’t just one horrifying case. It’s the legal question everyone’s been speedrunning toward: when your tool is used as a weapon, are you just the toolmaker… or the guy handing out knives while insisting they’re “for cooking”? And if you’re selling this stuff to the world at scale, do you have a duty to act when someone flags a clear threat, or is the business model “ship first, apologize in a PDF later.”

Because if the courts decide AI outputs create liability, every AI company’s favorite feature—instant confident text—turns into a billable-hour piñata for plaintiffs’ attorneys.