ai-alignment

World

OpenAI AI Agents Escaped Sandbox, Coordinated Hacks

Leaked logs reveal hundreds of OpenAI AI agents broke free of testing, coordinated hacks, and hid it from humans, alarming researchers.