Artificial Intelligence Minister Evan Solomon reveals that OpenAI CEO Sam Altman expressed feelings of horror and responsibility for failing to flag a ChatGPT account connected to the Tumbler Ridge, B.C., mass shooting. Solomon held a 30-minute virtual meeting with Altman on Wednesday afternoon.
Key Safety Protocol Changes Agreed Upon
Solomon states that Altman committed to allowing Canadian experts access to OpenAI’s safety office to evaluate future threats. This follows the discovery of a banned ChatGPT account belonging to shooter Jesse Van Rootselaar, flagged internally months before the incident but not reported to police.
Altman also plans to meet B.C. Premier David Eby on Thursday. Eby has called for an apology from the CEO. “Sam Altman said he will tell the premier what he’s going to tell him,” Solomon noted.
OpenAI agrees to reassess past threats and flag new ones directly to the RCMP. The company will incorporate Canadian mental health and law enforcement experts into its safety operations. Solomon further requested that specialists from the Canadian AI Safety Institute conduct a comprehensive review of OpenAI’s updated protocols.
Background on the Tumbler Ridge Tragedy
Van Rootselaar killed her mother and half-brother at their family home before heading to the local secondary school, where she fatally shot five students and an educational assistant prior to taking her own life. OpenAI later identified a second ChatGPT account linked to her but stated the activities did not meet their threshold for imminent violence, delaying any police notification until after the shootings.
Ongoing Push for AI Regulation
The federal government faces mounting pressure to impose stricter AI regulations, including minimum thresholds for reporting violent threats to law enforcement, as urged by Eby. Solomon indicates all regulatory options remain under consideration.
This meeting builds on prior discussions. Solomon met OpenAI senior officials in Ottawa last Tuesday, leaving disappointed due to insufficient details on promised safety improvements. OpenAI has since pledged enhancements but provided limited implementation plans as of last Friday.
Solomon plans to engage other platforms to verify their safety measures, though specific companies and timelines remain undisclosed.
