When should AI chatbots call the cops?

As more and more people divulge their innermost secrets to AI companions, platforms face the wrenching dilemma of determining when those admissions might signal a deadly threat to the public. Several Canadian ministers met with OpenAI representatives on February 25 to seek more information concerning why the company declined to notify authorities about the activities of a ChatGPT user who is the suspect in a recent mass shooting that devastated a rural town in British Columbia. The Wall Street Journal reported that OpenAI staffers had debated whether to go to the police after reviewing conversations in which the suspect described gun violence scenarios to ChatGPT. They ultimately decided to ban the user without reporting her. After the meeting, Canada’s AI minister described the decision by OpenAI employees as a “failure,” and warned that the government would impose more regulations on chatbots if companies don’t have appropriate public safety guardrails.