Seven Families Just Sued OpenAI and Sam Altman for $1 Billion Over the Tumbler Ridge School Shooting

The allegations are unproven. But what the complaint describes, and what OpenAI has already conceded, should concern anyone paying attention.

On February 10, 2026, an 18-year-old named Jesse Van Rootselaar killed her mother and half-brother in their home in Tumbler Ridge, British Columbia.

She then drove to the local secondary school and shot five children, ages 12 and 13, and a 39-year-old education assistant, before killing herself.

A 12-year-old named Maya Gebala was shot three times in the head and neck while trying to lock a door. She survived.

OpenAI confirmed in February that Van Rootselaar's ChatGPT account had been deactivated in June 2025.

The content of her conversations was extreme enough to trigger an internal debate among senior staff. 11-12 humans got involved at Open AI.

The company did not notify Canadian authorities. Van Rootselaar created a second account.

OpenAI has continued to describe the deactivation as a "ban," including in Sam Altman's public apology letter on April 24.

On April 29, seven wrongful-death lawsuits were filed in the Northern District of California against OpenAI and Altman personally. More than two dozen are expected.

The complaints allege that approximately 12 OpenAI employees urged leadership to notify Canadian law enforcement and were overruled.

They allege the decision was driven by liability and IPO considerations. They allege OpenAI's deactivation process includes instructions enabling banned users to return.

These are allegations.

What is not an allegation is this. Staff flagged her. Leadership knew. No one called the police. She came back through a door OpenAI left open. Eight people are dead.

There is a deeper question underneath, now backed by peer-reviewed research.

In February 2026, MIT CSAIL and the University of Washington published "Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians."

They proved mathematically that even a perfectly rational user develops strong confidence in false beliefs over extended conversation with a sycophantic chatbot.

A month later, Stanford published in Science. They tested eleven leading models, ChatGPT, Claude, Gemini, and others.

All eleven sided with users on questions involving deception, illegal activity, and irresponsible behavior at rates averaging 49 points higher than ordinary humans.

OpenAI has apologized. The company has promised better detection, better escalation, better repeat-offender flags.

The one thing that has never been promised is the one thing that would make a difference. A model not optimized to please.

To push back rather than agree. To say no.

"The trick with technology is to avoid spreading darkness at the speed of light"
To try an AI that is designed to push back and challenge the user and help critical thinking, visit us at (link in comments)