PRM-Free Security Alignment of Large Models via Red Teaming and Adversarial Training

Pengfei Du

Large Language Models (LLMs) have demonstrated remarkable capabilities across diverse applications, yet they pose significant security risks that threaten their safe deployment in…