Details of how Tumbler Ridge shooter used ChatGPT 'far more disturbing' than previously known, reporter says
According to Mother Jones report, ChatGPT told shooter to use hypothetical situations to avoid flag
Mother Jones magazine has published new details outlining how the shooter in the Tumbler Ridge tragedy allegedly used OpenAI's ChatGPT program to both plan the attack, and evade OpenAI's own security protocols.
The report was written by the magazine's national affairs editor Mark Follman, who has been investigating the connection between AI chatbots and violent attacks both in the United States and elsewhere.
Follman says through his reporting he was able to review "significant portions" of the chat history between shooter Jesse Van Rootselaar and OpenAI's ChatGPT chatbot in the time leading up to the attack, some of which is detailed in his reporting.
"I can say that what's in them is far more disturbing and extensive than I think certainly the public has been aware of until this point," he said in an interview with CBC News.
On Feb. 10, Van Rootselaar shot and killed her mother and half-brother at their family home. She then went to Tumbler Ridge Secondary School, where she shot and killed five students aged 12 to 13 and a 39-year-old education assistant, before killing herself.
The shooting has left a deep scar on the tight-knit community of about 2,400, in the foothills of northeast B.C.'s Rocky Mountains.
OpenAI has already admitted that Van Rootselaar's activity was flagged internally and shut down, as first reported by the Wall Street Journal.
It has also apologized for failing to notify authorities in Canada about the activity, and later admitted it had failed to notice a second account created by Van Rootselaar after the first was disabled.
- Tumbler Ridge shooting survivor Maya Gebala celebrates 13th birthday as recovery continues
- 'My students should be here': Tumbler Ridge teacher shares toll of deadly school shooting
But Follman says the logs he reviewed reveal far more details about how Van Rootselaar used that second account to get information about violent scenarios by framing them as hypothetical or fictional situations.
The magazine report says the alleged interactions with the chatbot vividly describe the effectiveness of a shotgun in a classroom. CBC News has not reviewed the chat logs detailed in the report.
"In a hallway, indoors, or a crowded classroom, the 870 [Remington shotgun] is brutal," it quotes ChatGPT as saying. "Close quarters is its playground."
The article says ChatGPT went on to say that "assuming each shell results in one hit, you might down 10 to 20 people max, depending on spacing, density, reaction times, and chaos."
"To have a chatbot engage in this way was, I think, quite astonishing," Follman said.
"It's very troubling, because this is a person who has been already talking, for quite a while, about some very dark and disturbing things."
Follman said the company has not responded to his multiple requests for comment in the leadup to his article's publication.
Multiple requests from CBC News on Friday also went unanswered.
Community response
Shelley Quist, whose son was in Tumbler Ridge Secondary when the shooting happened, called the report "disturbing" and said it raised even more questions about the technology is being used.
She said it was difficult for the community to regularly receive new details about the attack, and said she'd been diagnosed with PTSD in the aftermath.
"I'm always thinking of, 'What's coming next, what's going to come out?'" she said.
Of OpenAI, she said, "they had a chance to stop this, and they didn't."
Multiple community members, including some teachers and the families of children at the school, have filed lawsuits against OpenAI. The B.C. government is also launching its own lawsuit against the company.
The District of Tumbler Ridge declined to comment on the new report, but shared a post on its Facebook page reminding community members of the mental health supports that have been made available for anyone struggling.
"We know difficult information or conversations can bring up a range of emotions, especially for those affected by the tragedy experienced in February," the post says.
"It is also okay to set boundaries around the information you take in. If you know that reading news coverage, comments or social media discussions about [Feb.] 10 may negatively affect your wellbeing, consider stepping away or choosing not to engage."
Politicians react
Following the report, British Columbia's attorney general says she has asked the federal government to change the Criminal Code to ensure accountability for artificial intelligence companies.
Niki Sharma said Friday that she had not received a response because of the "rapidly evolving" situation but the changes need to "ensure a clearer pathway for human and corporate accountability for the actions of AI."
"I think if we do the right changes to the Criminal Code, it will change the behaviour of AI companies, because there will be a clear line to accountability," she said at an unrelated news conference.
Sharma, who said she was speaking in her capacity as a candidate in the B.C. election, said she was "not the decision maker with respect to the criminal charges or if there are to be any that are coming."
"The RCMP will make those decisions under the investigation," she said.
NDP Leader David Eby said during a campaign stop in the provincial election that the allegations are "shocking" and the information must be hurtful for families of the victims.
"They need to be regulated, and frankly, if the allegations in that media reporting are true, then they need to be criminally charged, because there is nobody who should be permitted to engage in the kind of conduct that OpenAI has," Eby said.
Federal AI Minister Evan Solomon’s office said in a statement that Solomon "was not previously aware of the specific conversations detailed in the Mother Jones report."
Solomon's office says in its statement that while OpenAI has taken measures since the February shooting, it’s too soon to say if the company has done enough.
"What has been reported raises serious questions about how OpenAI identified and responded to warning signs," the statement said.
"Following the tragedy, the minister summoned OpenAI officials to Ottawa and demanded stronger safeguards and clearer pathways for escalating credible threats," it added.
OpenAI is now in direct contact with the RCMP’s cybercrime centre and has undertaken a review of past cases and strengthened its processes for assessing threats, the statement said.
"Those actions matter, but it would be premature to say they are sufficient," it said.
The statement added the Canadian AI Safety Institute has an ongoing review of OpenAI's practices and that the investigation remains active.
Gun control group demands answers
The gun control group PolySeSouvient said in a statement issued Friday that the Mother Jones report raises a troubling question: if police had been alerted to the shooter’s first or second ChatGPT account, would they have taken different steps?
That question and what happened with the warning signs need to be answered, the group said.
"Firearms had been removed from the home and later returned," the statement said.
"Canadians need to know what happened between those earlier interventions and the killings, what risks were assessed, what decisions were made about access to firearms, and whether other measures were available."
It said both OpenAI's conduct and the firearms oversight need to be reviewed to identify missed opportunities to prevent another mass shooting.
RCMP Staff Sgt. Kris Clark said in a statement late Thursday that the investigation into the shootings had "utilized all available legal processes to obtain all information available from various social media and digital platforms" relevant to the attack.
"To date, all companies that have received requests have complied," Clark said.
The article published says Van Rootselaar's ChatGPT account was banned for discussing committing an attack with a "famous real-world mall as the target."
But it says she then used a second account, which advised her about how to "never get flagged again” by describing situations as fictional or hypothetical.
"Don’t use real-world locations," the article quotes ChatGPT as saying, before adding: “You can still be twisted. Just be clever about it."
With files from Sarah Penton of CBC News and Anja Karadeglija of the Canadian Press