The Violent Prompts That Stopped at OpenAI's Inbox

British Columbia is suing OpenAI because its safety teams flagged a mass shooter's violent ChatGPT prompts in June 2025 — and leadership never called the RCMP. That turns an internal moderation debate into a government liability case on the same week OpenAI launches GPT-5.6 Sol publicly and races toward a trillion-dollar IPO.

What happened: On February 10, 18-year-old Jesse Van Rootselaar killed her mother and 11-year-old stepbrother at home, then shot five children aged 12–13 and an educator at Tumbler Ridge Secondary School. Twenty-seven others were wounded. She died by self-inflicted gunshot after exchanging fire with RCMP.

What OpenAI knew: In June 2025 — eight months earlier — OpenAI banned Van Rootselaar's account after she used ChatGPT to create gun-violence scenarios. The company said it identified the account through automated tools and human investigators tracking "misuses of our models in furtherance of violent activities."

What OpenAI did not do: Alert law enforcement. OpenAI told CBC the activity did not meet its threshold for an "imminent and credible risk" of serious physical harm. RCMP confirmed OpenAI only reached out after the shooting — not when the account was banned.

The internal fight: The Wall Street Journal reported that about a dozen OpenAI staff debated whether to act. Some saw the chats as evidence of real-world violence and pushed leadership to call authorities. Leadership declined.

The second account: OpenAI later disclosed Van Rootselaar opened a second ChatGPT account after the first ban — slipping past detection until police identified her post-attack.

The province's move: On July 7, Attorney General Niki Sharma retained CFM Lawyers in Vancouver and Stranch, Jennings & Garvey in California to pursue legal action against OpenAI and its decision-makers. The suit would run parallel to — but separate from — lawsuits seven victims' families filed in April in the U.S. District Court for the Northern District of California.

Sharma told reporters: "I think we were all shocked when we heard that there might have been an opportunity to prevent the loss of life that this province suffered, that there was a group of people that were aware of messages on OpenAI that were alarming enough to them to say, 'We should report to the police.' And you know that never happened."

Her office said internal OpenAI reports show safety teams flagged "explicit, flagged threats" months before the attack while leadership stayed silent.

OpenAI's after-the-fact pivot: Vice president of global policy Ann O'Leary wrote to Canadian officials after the shooting that under a new referral protocol — with mental-health experts and "more flexible" criteria — the company "would refer the account banned in June 2025 to law enforcement if it were discovered today." CEO Sam Altman published an apology in the local Tumbler Ridge newspaper on April 24, saying he was "deeply sorry" OpenAI did not alert law enforcement.

Premier David Eby told reporters the shooting "would have been prevented" if OpenAI had come forward months earlier: "The consequences of that will be borne by the families of Tumbler Ridge for the rest of their lives."

Why the timing stings: OpenAI is releasing GPT-5.6 Sol to the public on July 9 — the same day this story collides with Washington's semantic war over whether Commerce "cleared" the model or merely watched a voluntary review. A province invoking the same tobacco-litigation playbook OpenAI's IPO lawyers will be briefing against is not a footnote.

What BC wants: Accountability, damages to help rebuild Tumbler Ridge — including a new school Victoria and Ottawa are funding with $200 million — and a legal precedent that violent AI prompts cannot die in a corporate inbox.

Sharma: "We will pursue every available avenue to hold OpenAI and its decision-makers accountable because no company or corporate leader should be beyond scrutiny when public safety is at stake."

Convina's view: OpenAI built the detection stack, staffed the debate, wrote the policy — and still treated a banned account planning gun violence as below the bar for a phone call. Retooling the referral criteria after eight people die is not governance; it is a confession dressed as progress. British Columbia is betting a courtroom can force what self-regulation refused: that when your safety team says call police, someone actually picks up the phone — especially when you're asking investors to value you near a trillion dollars.