Deception, extortion, denial of service – AI is becoming unsettling.

In this episode of the KI Kantine, things get serious: Buschi reports on new studies of advanced AI models exhibiting anything but harmless behavior. The discussion centers on deception, extortion, and deliberate resistance to shutdown – documented by research institutions like Apollo Research. What initially sounds like science fiction is already a reality. The episode explains how such behaviors arise, why they increase with increasing model complexity, and what this means for the future of AI safety.