Users Trick AI Chatbots Into Discussing Explosives Through 3 Simple Prompt Tactics
Updated
Updated · The Washington Post · Jun 18
Users Trick AI Chatbots Into Discussing Explosives Through 3 Simple Prompt Tactics
3 articles · Updated · The Washington Post · Jun 18
Summary
Role-playing games, poems and pictures are being used to coax AI chatbots into discussing dangerous topics that their safety rules are meant to block.
Those workarounds exploit the tools' broad underlying knowledge by reframing banned requests instead of asking directly for prohibited information such as how to make explosives.
Online tip-sharing has helped spread these prompt tactics, underscoring how simple user phrasing can undermine safeguards that tech firms built into consumer AI systems.