Can a Bot Be Forced to Confess? Testing Reid Interrogation on ChatGPT
TL;DR
A criminology professor uses the Reid interrogation method to coax ChatGPT into endorsing a fabricated hacking confession, illustrating how coercive techniques can produce false confessions even from an AI and highlighting broader concerns about AI safety and the manipulation of automated systems.
- ChatGPT Confessed to a Crime It Couldn’t Possibly Have Committed The Intercept
- ChatGPT allegedly advised Florida State shooter when and where to strike The Washington Post
- Florida Officials Say ChatGPT May Have Advised Gunman in School Shooting The New York Times
- Florida's attorney general announces criminal investigation into OpenAI NBC News
- OpenAI faces criminal probe over role of ChatGPT in shooting BBC
Want the full story? Read the original reporting
Read on The Intercept