Red Team Diffuser: Exposing Toxic Continuation Vulnerabilities in Vision-Language Models via Reinforcement Learning

VLMRe-assign community