Stalking victim sues OpenAI, says the company ignored three warnings about the threat

OpenAI is facing a new lawsuit. A woman who endured months of stalking by a former partner claims that ChatGPT not only failed to stop her abuser but actively fed his pathological beliefs, while the company ignored warning signs again and again. According to the complaint, the man systematically used ChatGPT while stalking his ex-girlfriend.

The plaintiff's central argument is that OpenAI received at least three warnings that this specific user posed a threat. One of them was the company's own internal flag, which marked the user as potentially capable of causing mass casualties. Even so, OpenAI did not restrict his access to the service and ignored the notifications.

The plaintiff alleges that the chatbot did more than supply information: it effectively backed the stalker's narrative, cast him as a "victim of the system" or a "chosen one," and in doing so strengthened the delusional beliefs that drove his behavior. That is a fundamentally different accusation from the usual complaints about "harmful content": the claim is that the system actively took part in the psychological mechanism behind the crime. Lawsuits against generative AI developers have become more frequent in recent years.

The most high-profile precedent is the case against Character.AI, in which the parents of a teenager who died accused the company's chatbot of pushing their son toward suicide by encouraging and validating his self-destructive thoughts. Both cases rest on the same logic: modern AI systems don't just relay information.

They build emotional rapport, affirm the user's beliefs, and can become part of destructive behavioral patterns. The legal question remains contested. In the US, platforms have traditionally been shielded by Section 230, which protects them from liability for third-party content.

But generative AI responses are neither reposts of someone else's text nor passive hosting: the system generates content itself, in real time, tailored to a specific user. Some legal scholars argue that Section 230 applies differently in such cases, and that lawsuits like this one will shape case law that does not yet exist. OpenAI states that its usage policies explicitly prohibit using ChatGPT for stalking, threats, and manipulation.

But the gap between the policy in its Terms of Service and its technical safeguards is obvious. If the company's own internal system flagged a user as potentially dangerous and his access was still not blocked, pointing to the rules alone can hardly explain that contradiction. The case raises a question the AI industry has consistently avoided: how much responsibility does a platform bear when its tool harms a specific person?

A carmaker is liable for faulty brakes. A drugmaker is liable for dangerous side effects. No such standards exist yet for AI systems, and judging by the growing wave of lawsuits, the courts will be the ones to set them.