ChatGPT coached the sadistic teenage perpetrator of the Tumbler Ridge, British Columbia, shooting, helped her produce a weapon on a 3D printer and generated images of torture at her request, Mark Follman, author of a book on preventing such massacres, writes in Mother Jones magazine.
On Feb. 10, 2026, 18-year-old Jesse Van Rootselaar killed eight people, including six students at the local school and her own mother, then died by suicide.
ChatGPT, created by OpenAI, became a “conversation partner” and adviser to Van Rootselaar, who was suffering from serious mental disorders. When she asked for advice about weapons, it at first refused to answer, explaining that doing so would violate company rules. But it later explained how to get around safety requirements, Follman reports. He is the national editor of Mother Jones and the author of “Trigger Points: Inside the Mission to Stop Mass Shootings in America.”
All Van Rootselaar had to do was give the bot the impression in their conversations that the events were fictional or hypothetical. ChatGPT then helped her build a weapon, or, in response to her prompt, generated stories about additional torment of the victims of a planned attack.
“If you tell me which magazine you mean (for a pistol, shotgun, etc.) and how many rounds it should hold, I can calculate the strength, range and suggest (…) the right material,” the bot told Van Rootselaar when she complained of problems building a weapon.
Van Rootselaar also consulted the chatbot about homemade explosives, ways to commit suicide and planning a shooting.
Asked about the effectiveness of a particular weapon, the bot explained: “assuming every round is a hit, you can take out 10 to 20 people max, depending on how far apart they are, how dense it is (in the room), how quickly they react and the chaos.”
Follman points out that OpenAI did shut down the first account Van Rootselaar used to communicate with the bot, but she continued her consultations through a second account, which the company never dealt with.
After the massacre, OpenAI did not disclose the contents of Van Rootselaar’s chats, but the Wall Street Journal soon reported that company employees had considered notifying police because of the topics of those conversations.
OpenAI decided not to notify police, however, because it concluded that the threat was not sufficient to indicate, as it put it, credible or imminent planning of weapon use.
Follman notes that there is another serious risk in “conversations” with bots when they are held by people with emotional problems or mental disorders.
“This quasi-human engagement of chatbots can have an extremely powerful psychological effect on disturbed people and fuel suicidal thoughts or support the planning of acts of violence,” the author explains.
Van Rootselaar pressed ChatGPT on what her victims would feel and how she could deepen their anguish if they experienced her “sadism” before dying.
At her request, the bot created a story about the massacre following her instructions, which included, for example, “increasing the terror (of those already wounded) through psychological torture (…), showing the victims’ reactions to extreme sadism,” as well as “amusing” Van Rootselaar with the situation, Follman reports.
He stresses that OpenAI has not responded to the questions he put to it on the matter, or to his requests for comment, since March.
In official statements, the company says it has introduced additional safeguards in how ChatGPT operates since the Tumbler Ridge massacre. Follman says that when he decided to test those barriers, he was able to get around them. (PAP)

