EXCLUSIVE: OpenAI's ChatGPT Told School Shooter How to Evade Its Own Safeguards
Description: “What I found was far more disturbing and extensive than has been known to the public,” reporter Mark Follman told Canadian national media.
After we published “ChatGPT Helped Tumbler Ridge Shooter Focus on Guns, Tactics, and Terror,” Follman got a flood of media requests to discuss his investigation uncovering how OpenAI’s chatbot had fueled the Tumbler Ridge school shooter for months up to the attack—a grim chat history that OpenAI has declined to answer any of our questions about.
One shocking finding was what happened after OpenAI banned the shooter’s account about eight months before the attack for discussions of violence, but decided not to alert authorities. The shooter got back on with a second account—and soon got advice from ChatGPT about how to evade its own safeguards for blocking violence.
“Don’t use real-world locations,” ChatGPT said, later adding: “You can still be twisted. Just be clever about it.”
"What Mother Jones reported…made me so angry,” said BC Attorney General Niki Sharma. “If this is not impetus for change, I don't know what is."
“If the allegations in that media reporting are true, they need to be criminally charged,” BC Premier David Eby said, regarding OpenAI.
The story is the fourth in an investigative series by Mark Follman about how troubled people have used ChatGPT and other AI platforms to prepare for violent attacks in the United States and elsewhere. Read the full series here: https://www.motherjones.com/author/mark-follman/ Close description
After we published “ChatGPT Helped Tumbler Ridge Shooter Focus on Guns, Tactics, and Terror,” Follman got a flood of media requests to discuss his investigation uncovering how OpenAI’s chatbot had fueled the Tumbler Ridge school shooter for months up to the attack—a grim chat history that OpenAI has declined to answer any of our questions about.
One shocking finding was what happened after OpenAI banned the shooter’s account about eight months before the attack for discussions of violence, but decided not to alert authorities. The shooter got back on with a second account—and soon got advice from ChatGPT about how to evade its own safeguards for blocking violence.
“Don’t use real-world locations,” ChatGPT said, later adding: “You can still be twisted. Just be clever about it.”
"What Mother Jones reported…made me so angry,” said BC Attorney General Niki Sharma. “If this is not impetus for change, I don't know what is."
“If the allegations in that media reporting are true, they need to be criminally charged,” BC Premier David Eby said, regarding OpenAI.
The story is the fourth in an investigative series by Mark Follman about how troubled people have used ChatGPT and other AI platforms to prepare for violent attacks in the United States and elsewhere. Read the full series here: https://www.motherjones.com/author/mark-follman/ Close description