A Red Teaming Roadmap Towards System-Level Safety

Zifan Wang, Christina Q. Knight, Jeremy Kritz +2

Large Language Model (LLM) safeguards, which implement request refusals, have become a widely adopted mitigation strategy against misuse. At the intersection of adversarial machine…