Diffusion LLMs as Targets and Adversaries: Mechanistic Safety Exploits

Part of

Self-Driving Car Security Evaluation