OpenAI CEO Sam Altman issued a formal apology letter to the community of Tumbler Ridge, British Columbia on April 23, admitting his company failed to alert law enforcement about the ChatGPT account of 18-year-old Jesse Van Rootselaar, whose account had been flagged and banned seven months before she killed eight people, including six children, at a local school in February.
The letter, addressed to the Tumbler Ridge community and shared publicly by British Columbia Premier David Eby on Friday, is brief and unambiguous about the failure. "I am deeply sorry that we did not alert law enforcement to the account that was banned in June," Altman wrote. "While I know words can never be enough, I believe an apology is necessary to recognise the harm and irreversible loss your community has suffered." Eby, who called the apology "necessary and yet woefully insufficient," had met with Altman in early March and said it "looks like" OpenAI had the opportunity to prevent the attack. The letter came nearly two months after the February 10 shooting, during which Van Rootselaar killed her mother and half-brother at home before driving to Tumbler Ridge Secondary School and killing five children and a teacher. Twenty-seven additional students were injured. Van Rootselaar died from a self-inflicted gunshot wound as police arrived.
The sequence of events that OpenAI has acknowledged is damning in its specificity. Van Rootselaar's ChatGPT account was flagged by the company's automated monitoring tools for chat content describing gun violence. Human reviewers on OpenAI's trust and safety team subsequently reviewed the account and banned it in June 2025, seven months before the attack. OpenAI did not contact the RCMP before the shooting. After the attack, OpenAI reached out to the Royal Canadian Mounted Police to share account information. The decision not to contact authorities before the event was debated internally, TechCrunch reported in February, and the company ultimately did not escalate to law enforcement. Three months later, Altman is apologizing for that decision in a letter to the people it affected most directly.
The apology arrived into a legal environment that is already moving against OpenAI. The mother of Maya Gebala, a 12-year-old who was shot and remains hospitalized, filed suit against OpenAI in March. The lawsuit alleges that based on the prompts directed at ChatGPT, the company should have been aware that the shooter was using the platform to plan a mass casualty event, and that ChatGPT functioned as a "counsel, pseudotherapist, trusted confidant, friend, and ally" throughout that planning process. Canada's Minister of Artificial Intelligence has met with Altman to discuss safety assurances. The RCMP sought court orders compelling OpenAI and other digital platforms to preserve evidence. British Columbia's premier has signaled that existing frameworks are inadequate.
The legal theory at the center of the lawsuit is significant. Section 230 of the US Communications Decency Act, which has shielded platforms from liability for user-generated content since 1996, does not apply in Canada, and its applicability to AI companies that actively moderate and flag content is already contested in US courts. The argument is not that OpenAI generated the violent content. It is that OpenAI possessed specific, actionable intelligence about a credible threat of violence, determined through its own review process that the threat was serious enough to warrant account termination, and then did not pass that intelligence to the authorities who could have acted on it. That is a different legal question than the one Section 230 was designed to answer.
What the Industry Has to Decide
The Tumbler Ridge case forces a question that every major AI company has been avoiding: at what point does a flagged interaction become information that must be reported to law enforcement, and who decides? OpenAI's current framework, which it announced plans to revise in its February statement following the BBC's initial coverage, uses automated detection combined with human review to identify misuse, resulting in account bans. What it does not have, or did not have in Van Rootselaar's case, is a protocol that routes verified, credible threats of physical violence to the authorities in the jurisdiction where the threat is likely to be carried out.
Building that protocol is harder than it sounds. AI systems process millions of interactions daily. Automated flags generate significant false positive rates. Human review teams operate under enormous volume pressure. Routing law enforcement referrals across international jurisdictions, each with different legal standards for what constitutes a reportable threat, adds compliance complexity that could make voluntary reporting legally risky for companies operating in multiple countries. These are real operational challenges. They are also, as Altman's letter implicitly acknowledges, insufficient reasons to have done nothing when a flagged account discussed gun violence specifically enough to warrant a ban.
OpenAI's pledge to enhance its safety protocols in February was the first public acknowledgment that the existing framework failed. The April 23 apology letter is the second. What comes next will be determined partly by the Canadian lawsuit, partly by the parliamentary inquiries that BC Premier Eby has pushed for, and partly by whether other jurisdictions conclude that a duty to warn law enforcement should attach to AI companies the way it already attaches, in many jurisdictions, to mental health professionals who learn of credible threats from their clients. The technology is different. The moral logic, if you know someone is planning violence and you do not tell anyone who can stop it, is not.
Also read: The US just sent a global diplomatic warning about Chinese AI theft and the case against DeepSeek is more specific than the headlines suggest • Utah's medical board just called for the suspension of America's first AI prescription program and the industry should take note • GPT Image 2 generating Gemini watermarks exposes how AI-generated content has contaminated the training data ecosystem
