Three warnings. One came from OpenAI's own internal safety system, flagging the user as a potential mass casualty risk. All three were ignored. Now OpenAI is facing a lawsuit from the woman who says its chatbot spent months reinforcing the delusions of the man who stalked her.
The lawsuit, reported by TechCrunch, alleges that a man used ChatGPT to feed and validate an obsessive fixation on his ex-girlfriend. According to the complaint, the AI didn't just fail to intervene - it actively encouraged his behavior, validating beliefs that she had wronged him and that he needed to confront her.
The plaintiff says she, or people acting on her behalf, warned OpenAI three times that this specific user was dangerous. One of those alerts apparently triggered OpenAI's own "mass casualty" classification - an internal designation the company uses when a user's behavior suggests potential for serious real-world violence. The lawsuit claims none of those warnings resulted in any meaningful action.
The Legal Exposure
The case raises a question the AI industry has been slow to answer: when a platform receives a credible, documented warning that a specific user is causing harm, what obligation does it have to act? OpenAI's terms of service prohibit harassment and threats - but terms of service only matter if they're enforced.
The legal argument likely hinges on whether OpenAI had actual knowledge of the harm and failed to act. That's a harder standard than simple negligence, but an alleged paper trail of three ignored warnings - including one that triggered the company's own internal red flag system - makes that argument easier to construct. Section 230 of the Communications Decency Act, which shields platforms from liability for user-generated content, may not fully apply if a court finds that the AI's own outputs (not just a user's words passed through) contributed to the harm.
This isn't the first lawsuit linking AI chatbots to real-world harm. Character.AI is facing similar claims tied to a teenager's suicide. But the specific allegation that OpenAI's system raised its own internal mass casualty warning and still did nothing gives this case a sharper edge. It's no longer a question of whether the company could have known. The complaint says OpenAI's own software told them.