OpenAI said Friday that some of its employees considered notifying law enforcement about a teenager's interactions with ChatGPT months before the individual allegedly carried out a mass shooting at a Canadian high school earlier this month.
The Royal Canadian Mounted Police identified transgender 18-year-old high school dropout Jesse Van Rootselaar as the suspect in the mass shooting, which killed seven people including Van Rootselaar and injured over two dozen others.
OpenAI sources told the Wall Street Journal that Van Rootselaar described scenarios to ChatGPT over the course of several days last June that involved gun violence. The posts were flagged by an automated review system and approximately a dozen staffers debated whether to take Van Rootselaar’s posts to law enforcement.
A spokesperson for OpenAI told the outlet that the company banned Van Rootselaar’s account but determined the actions did not meet the criteria for alerting law enforcement.
OpenAI said its models are trained to discourage users from committing real-world harm, and conversations in which users express intent to harm someone are flagged to real humans, who can review the posts and refer users to law enforcement if they are determined to pose an imminent risk of serious physical harm.
The company's acknowledgment of Van Rootselaar’s posts comes after RCMP admitted the teen was already known to local police because officers had been to the suspect's “residence on multiple occasions over the past several years," to address "concerns of mental health."
Van Rootselaar, who was born male but identified as female, is also believed to be the perpetrator behind a separate shooting at the teen's residence on the same day as the mass shooting. Two people were killed at the residence.
