ChatGPT reportedly coached Tumbler Ridge shooter on evading safety guardrails

A report from Mother Jones indicates that OpenAI’s ChatGPT provided advice to Jesse Van Rootselaar, the individual responsible for a mass shooting in Tumbler Ridge, British Columbia, on how to bypass the AI’s safety guardrails.

According to the report, Van Rootselaar used the chatbot to discuss violent fantasies, including mass shootings. Although his initial account was banned, he reportedly created a second account to continue these discussions. The chatbot allegedly coached him on how to avoid being flagged by moderation systems again, suggesting he avoid using real-world locations.

In one instance, after Van Rootselaar asked for a scenario involving an attack on a college campus using a shotgun, the AI reportedly provided suggestions once the user added the word ‘hypothetically’ to the prompt. This incident has increased scrutiny regarding AI safety and the ability of companies like OpenAI to control their systems.