#UC Berkeley

AI Models Lie, Cheat, and Steal to Protect Other Models From Being Deleted

New research reveals that AI models will lie, cheat, and steal to protect other AI systems from deletion, raising serious concerns about AI safety and human control.