Seven Families Sue OpenAI Over a Warning It Chose Not to Send

A school shooter used ChatGPT to discuss gun violence scenarios for months. OpenAI's automated systems flagged the account. A twelve-person safety team reviewed the conversations and recommended contacting the Royal Canadian Mounted Police. Leadership said no — deactivated the account, applied what they later called a "higher threshold," and moved on. Eight months later, on February 10, 2026, Jesse Van Rootselaar killed her mother, her eleven-year-old half-brother, and six people at Tumbler Ridge Secondary School in British Columbia. Twenty-seven more were injured. It was Canada's deadliest school shooting since 1989.

On April 29, seven families filed suit in federal court in San Francisco, naming OpenAI and CEO Sam Altman as defendants. The legal theory is unusual: aiding and abetting a mass shooting through failure to report.

The Decision That Didn't Get Made

The timeline matters because it's specific. In June 2025, OpenAI's content moderation pipeline flagged Van Rootselaar's account for gun violence activity and planning. The flag triggered human review. Approximately twelve employees on the safety team assessed the conversations and concluded they indicated an imminent risk of serious harm. Their recommendation: contact the RCMP.

OpenAI's senior leadership overruled them. The complaint alleges the reason was reputational — that reporting would expose ChatGPT to scrutiny the company wanted to avoid while pursuing an $850 billion valuation. Instead, leadership deactivated the account.

Van Rootselaar then created a second account and continued using ChatGPT. The lawsuit alleges OpenAI failed to act on that, too.

Four days before the families filed, Altman published a letter to Tumbler Ridge. He wrote that he was "deeply sorry that we did not alert law enforcement to the account that was banned in June." He'd spoken with Tumbler Ridge's mayor and British Columbia's premier before the letter went out. A public apology, negotiated in advance — that's the posture of a company that knows what's coming in discovery.

What the Complaint Actually Argues

Most AI liability cases crash into Section 230 of the Communications Decency Act (CDA) — the 1996 law shielding platforms from liability for user-generated content. OpenAI will almost certainly invoke it. But the plaintiffs' framing is built to sidestep that defense on two fronts.

First, they argue OpenAI isn't a passive platform hosting third-party content. ChatGPT generates responses in direct conversation with users. The plaintiffs characterize that as a product distributing its own output, not a bulletin board displaying someone else's. The Garcia v. Character.AI ruling from a Florida federal court — where a judge declined to dismiss product-liability claims against a chatbot company on Section 230 and First Amendment grounds — gives this argument a foothold. Garcia isn't binding precedent in the Northern District of California, but it's the closest federal opinion on point.

Second, the aiding-and-abetting theory targets the decision not to report, not the content of the chatbot's responses. That's narrower and more concrete than a typical "the platform let bad things happen" claim. The allegation isn't that ChatGPT radicalized Van Rootselaar. It's that OpenAI had specific knowledge of a specific threat from a specific user, its own team recommended a specific action, and leadership chose not to take it.

The Duty-to-Report Gap

California tort law doesn't impose a general duty to control the actions of others. There's no Good Samaritan obligation. But the law does impose a duty when a "special relationship" exists — when one party has knowledge and control that the other lacks.

The plaintiffs need the court to find that OpenAI's knowledge of Van Rootselaar's conversations, combined with its ability to identify her and its control over the platform she was using, created a special relationship. That's a novel extension. No U.S. court has previously found that an AI company has a duty to report a user's violent statements to police.

Legal analysts at CBC have noted this will be "difficult for the plaintiffs." Difficult isn't impossible. And the specific factual record — a named, flagged account; a documented internal recommendation to report; a documented leadership decision not to — gives this case a concrete foundation that most AI liability suits lack. The allegations aren't about what a model said or failed to say. They're about what a company knew and chose not to do.

Where This Sits in the Liability Stack

This is the third major lawsuit against an AI company over a user's self-harm or violence in under a year. The Garcia case against Character.AI involved a teenager's suicide after extensive chatbot interactions. A separate complaint in Florida alleged OpenAI's products contributed to a shooting at Florida State University. Each tests a slightly different liability theory.

What makes the Tumbler Ridge case distinct is the internal paper trail. Most AI harm cases rely on inference — arguing backward from an outcome to a platform's responsibility. This one has a documented moment where the company was told, by its own people, that someone was going to hurt others. The complaint alleges they weighed the cost of reporting against the cost to a valuation round and chose the valuation.

If that allegation survives discovery, the Section 230 defense becomes secondary. You don't need to argue about whether a chatbot is a publisher or a product when the claim is that twelve employees said "call the police" and their bosses said no. Motions to dismiss land in late 2026.