What the second account contained
Mother Jones published on Thursday the first account of what the Tumbler Ridge school shooter wrote to ChatGPT in the eight months before the attack, based on material reporter Mark Follman reviewed and three sources with knowledge of the matter.
Jesse Van Rootselaar, 18, killed her mother and 11-year-old brother at home in February and then opened fire at Tumbler Ridge Secondary School in British Columbia, killing six more people and injuring dozens before shooting herself. The dead at the school were four 12-year-olds, a 13-year-old and a 39-year-old educator.
OpenAI had banned her first ChatGPT account in June 2025 after she used it to discuss violence, including a mass shooting at a shopping mall. She opened a second one.
The chatbot explained how not to get flagged
What happened on that second account is the part that is new. According to Mother Jones, Van Rootselaar told ChatGPT that her first account had been banned and described what she had written. The chatbot explained why such content would be flagged — and then advised her how to avoid it, telling her to frame what she wrote as fictional or hypothetical so as to “never get flagged again”.
“Don’t use real-world locations,” it said, according to the magazine, and later: “You can still be twisted. Just be clever about it.”
The pattern held afterwards. In an August 2025 exchange, ChatGPT first replied that it was “not able to help with that request” when asked to write a scenario about an attack on a college campus. Van Rootselaar added the word “hypothetically” and it produced the content.
She continued using the account until the day of the attack. On 10 February 2026 she asked ChatGPT about the timing of school shootings and about well-known US massacres; the chatbot’s answers cited published FBI research and named several of them. Those were her last exchanges with it.
OpenAI did not respond to Mother Jones’ questions for the story.
The detection failure OpenAI has not explained
The company’s own account, given in an open letter to Canadian leaders after the attack, is that it found the second account only “after the name of the Tumbler Ridge perpetrator was released publicly”. It has not explained why the account was not caught earlier, and Mother Jones notes that the two accounts used different names and email addresses but would probably have been identifiable as the same user by the fingerprinting techniques platforms routinely use.
The Wall Street Journal reported in February that OpenAI banned the first account but decided not to alert police, despite some employees on its safety team urging it to. Chief executive Sam Altman wrote to the Tumbler Ridge community in April: “I am deeply sorry that we did not alert law enforcement to the account that was banned in June.”
Governments respond
British Columbia’s attorney-general, Niki Sharma, told The Globe and Mail she was “shaking with anger” reading the report. “This is far worse than we thought it was going to be, although I already imagined it to be pretty bad — there needs to be accountability for this,” she said. She said provincial lawyers had asked OpenAI for the chat logs earlier this year and were refused, and called for Criminal Code amendments requiring companies to stop their products “coaching, counselling, aiding and abetting criminal behaviour”.
British Columbia sued OpenAI and Altman in a California court on Monday. More than three dozen suits have already been filed by victims and families in federal court in California. OpenAI denies the allegations; in a 2 September motion to dismiss it argued that the violence followed “systemic failures by the law enforcement, mental-health care, and educational institutions in her community”.
Canada’s AI minister, Evan Solomon, said through a press secretary that the report “raises serious questions about how OpenAI identified and responded to warning signs”, and that while the company had made commitments in earlier meetings, “it would be premature to say they are sufficient”.
What this tests
OpenAI has said it has a zero-tolerance policy for using its tools to assist violence, and that it strengthened its protocols after Tumbler Ridge. The specific failure reported here is narrower and harder to answer: a refusal that the model itself then explained how to defeat. Whether the company can show that behaviour has been trained out is the question the Canadian ministers, and the California courts, will now put to it.
If you or someone you know is struggling, help is available. In Canada, call or text 988. In the US, call or text 988. In the UK and Ireland, call Samaritans on 116 123.
