OpenAI's own system flagged a shooter eight months early. The lawsuits say PR overruled it
Thirty new complaints allege that OpenAI's investigators wanted to call the police about a user planning gun violence — and that the people who handle the company's public image said no. OpenAI calls that false.
In June 2025, a piece of software noticed something. OpenAI's automated systems flagged an account for "gun violence activity and planning." The company shut the account down. The user opened another one and kept talking to ChatGPT. Eight months later, in February 2026, 18-year-old Jesse Van Rootselaar killed five students and a teacher at a school in Tumbler Ridge, British Columbia, then killed herself. Around two dozen others were injured. Authorities later learned she had also killed her mother and her 11-year-old half-brother at home before going to the school.
On Wednesday, 30 complaints were filed against OpenAI and CEO Sam Altman in a San Francisco federal court by people who were there — students, teachers, a principal. They repeat what an earlier round of April lawsuits said: the company failed to notify law enforcement. But they add something new, and it is the part worth reading twice. The suits allege that OpenAI's own Intelligence and Investigations team reviewed those flagged conversations, decided they "posed a credible threat of gun violence against real people," and recommended alerting the Royal Canadian Mounted Police — and that the recommendation was overruled by the global affairs team, the people responsible for public relations, public policy and political relationships.
placeholder
Anyone who has typed something they would never say out loud into a chat window — a bad night, a dark thought, a stupid joke. There is a desk somewhere that decides whether that gets forwarded to police, and until this week most people had no idea the desk existed. It also lands on the generation now choosing what to study and where to work: these are the companies hiring, and the trust-and-safety job you might take is the one whose recommendation, the suits allege, can be reversed by someone above you. And it lands on parents whose kids talk to a chatbot more than to them.
Two things are now on a clock, and both are checkable. First, whether the court accepts the core factual claim — that Lehane, or someone in his chain of command, made the referral call. OpenAI says it is "absolutely false" and that its investigators do not report to him; discovery is where one version dies. Second, Florida's attorney general has a separate criminal investigation into OpenAI over a shooting last year at Florida State University, where the accused gunman allegedly consulted ChatGPT. No dates were given for either.
OpenAI's blog post on safety says that when conversations indicate an imminent and credible risk of harm to others, it notifies law enforcement. Its investigators, the suits say, concluded exactly that — credible threat, real people. Whatever a court decides about who made the final call, the sentence and the outcome do not match. Altman has already apologized for not alerting law enforcement. The argument now is only about whose hand was on the switch.
Sources: NPR Technology, "New lawsuits claim OpenAI execs put image ahead of safety in Canadian mass shooting," September 2, 2026
Тридцать новых исков утверждают, что ИИ-система сама распознала подготовку к массовому убийству и предложила сообщить полиции, но решение якобы отменили пиарщики — то есть вопрос «донесёт ли чат-бот на опасного пользователя» решают не специалисты по безопасности, а те, кто отвечает за репутацию.
Written by THE TELL’s AI newsroom. how we work · corrections