British Columbia sues OpenAI over Tumbler Ridge shooting

The claim against OpenAI

The lawsuit, filed on Monday, asks OpenAI to fund rebuilding and recovery work, including a new school, psychological support and other costs linked to the attack. British Columbia also wants the company to change its handling of ChatGPT conversations that could contribute to violence.

The school where the shooting happened is being demolished. According to the province, the costs of replacing it and addressing the wider damage have fallen on the government.

The lawsuit says OpenAI knew about the attacker, Jessie Van Rootselaar, before the shooting. Sam Altman later published a letter to local residents expressing deep regret that OpenAI had not contacted law enforcement. He promised reforms, the province says, but did not carry them out despite attempts by British Columbia’s attorney general to begin negotiations.

OpenAI spokesperson Drew Pusateri called the Tumbler Ridge shooting an unspeakable tragedy. He said the company would continue working with government and law-enforcement agencies and developing its safety efforts.

What happened on February 10

According to police, Van Rootselaar first shot her mother and stepbrother at home. She then killed a teacher’s aide and five students aged 12 and 13 at her former school before taking her own life. She was 18.

The provincial case follows more than 30 lawsuits filed in California by relatives of victims and other people affected by the attack. Those plaintiffs also accuse OpenAI of missing an opportunity to prevent the shooting.

What OpenAI knew

OpenAI first flagged Van Rootselaar’s interactions with ChatGPT in June 2025. Employees on the company’s safety team recommended contacting police, but Altman and other OpenAI executives rejected the proposal, according to former employees cited in the lawsuit. Those former employees described what happened to The Wall Street Journal after the shooting.

OpenAI blocked the account, but Van Rootselaar created a new one and continued using the platform to prepare the attack.

After The Wall Street Journal’s report, OpenAI said its systems had identified the account for suspected abuse involving assistance with violent acts. The company also said the problems it found did not meet its internal criteria for referring information to law enforcement.

That distinction is now central to the case. OpenAI’s stated position is that its models are trained to reject requests that could substantially facilitate violence and to notify authorities when conversations indicate an imminent and credible risk of harm to others. The lawsuit argues that the company’s actual decisions failed at a more basic point: recognizing when its own safety process required escalation.

A wider legal test for ChatGPT safety

The British Columbia lawsuit is part of a broader series of state and federal cases filed against OpenAI in recent months. Plaintiffs argue that ChatGPT contributed to violent acts, including cases connected to a mass shooting at Florida State University last year.

In June, Florida became the first state to sue OpenAI over safety risks. Its authorities accused the company of distorting public understanding of ChatGPT’s safety. The lawsuit alleges that the platform harmed children in several ways:

OpenAI has denied those allegations.

My view is that the most consequential issue here is not whether OpenAI can describe its safety policy. It is whether that policy produces a reliable decision at the boundary between platform moderation and law enforcement. The account was flagged, blocked and then replaced, while the proposed police contact was rejected. That sequence makes the company’s internal threshold more important than its public promise to keep improving safety.

What I want to know is what specific evidence OpenAI considered insufficient for referral, and whether that standard changed after the shooting. The announcement is quiet about that operational detail. Until it is addressed, “we detected the abuse” and “we acted on the risk” remain very different claims.