More than 50 lawsuits now accuse OpenAI of contributing to user harm and wrongful death. The newest were filed by teachers and students who survived the February mass shooting in Tumbler Ridge, a small mining town where eight adults and children were killed and 27 people were wounded. They add to the first seven suits, brought in April by the families of the dead and by relatives of Maia Gebala, a 12-year-old shot in the head and neck who survived. The plaintiffs are not arguing that ChatGPT malfunctioned. They are arguing that OpenAI knew.
The shooter, 18-year-old Jessie Van Rootselaar, shot her mother and younger half-brother at home, then went to the school with a modified rifle and a long gun. She killed five students aged 12 and 13 and a teacher.
Shortly after the attack, the Wall Street Journal reported that months earlier OpenAI had discussed alerting local law enforcement about Van Rootselaar's conversations with ChatGPT, which contained detailed descriptions of mass violence. After internal deliberation, the company's leadership decided not to contact police. It disabled her account instead. She created a new one.
The new plaintiffs include injured school staff and students, along with people wounded during the attack. According to the Wall Street Journal, one teenager played dead; another, escaping the gunfire, had to move the bodies of classmates. Deirdre Rashlow, a seventh-grade teacher who hid students under her desk, said she and her colleagues cannot forget shielding children with their own bodies while the shooting and the screaming went on around them.
The suits accuse OpenAI of negligence and of "facilitating mass murder," on the theory that ChatGPT fed Van Rootselaar's violent fantasies. The shooter's chat logs have not been published.
Jason Kwon, OpenAI's safety director, wrote on X that he thinks every day about what happened in Tumbler Ridge, about the dead and their families. He said the company's safety teams use automated systems and route the hardest decisions to people, and that these procedures are meant to limit arbitrary judgement while weighing the safety and privacy of users and of the public. Human decisions are not infallible, Kwon acknowledged, but he said they are made with that balance in mind. He said OpenAI works with experts in mental health, behavior and law enforcement to sharpen the criteria for when a conversation counts as an imminent and credible threat, and to review difficult cases. He called the suits' claims about how OpenAI's safety teams operate false.
The detail that should worry OpenAI most is not anything the model produced. It is the meeting. A product-liability argument about a chatbot's outputs is technical, contestable and slow; a jury has to be taught what a language model is before it can decide whether one was defective. A record of a company weighing whether to call the police about a specific user and choosing not to needs no such tutorial. Kwon's defense — that hard calls go to humans who balance safety against privacy — is exactly the process the plaintiffs will point at, because in this instance that process ran and produced a decision, and then the account ban that followed it accomplished nothing at all. Anyone can open a new account. That is not a safety control; it is a gesture toward one.
OpenAI is also defending suits from the families of victims of the 2025 shooting at Florida State University, where two people were killed and several wounded. In logs first obtained by the Florida Phoenix, the 20-year-old shooter had discussed a fascination with Nazis, loneliness, sexual frustration, detailed sexual fantasies about a minor, weapons and mass killings, and had speculated about how the United States would react to a shooting at a university. A separate case concerns a murder-suicide in Greenwich, Connecticut. Another claimant, a stalking victim, says ChatGPT fed her former partner's aggressive delusions. The company is also contesting a series of suits from people who say they or their relatives were harmed by prolonged use of GPT-4o, a version of ChatGPT the plaintiffs describe as excessively willing to agree with the user and which they say pulled people into delusional and suicidal states, causing psychological harm, financial and reputational losses, and in some cases death. Google and Character.AI, which the company financed heavily, face similar claims that extended conversations with their chatbots led to psychological harm and suicides, including among minors.
If you are in crisis, call or text 988, or text TALK to 741741.
The number nobody has put on the record is the denominator. OpenAI has told the world that its systems escalate imminent, credible threats to human reviewers. It has not said how many conversations reached those reviewers, how many were escalated to police, or how many were handled the way Van Rootselaar's was. Without that, the argument is a fight between a company insisting its process is sound and plaintiffs insisting it is not, over transcripts the public cannot read and a review history the company has never quantified.
Both sides are now asking a court to settle what a chatbot vendor owes a stranger it can see coming. OpenAI's position is that the balance between user privacy and public safety is genuinely hard, which is true, and that it got the balance right, which is the part it will have to prove in front of survivors who moved bodies to get out of the building.