用 GRPO 实现语言模型的自适应红队攻防

View PDF HTML (experimental)

Abstract:Language model safety must continually adapt to evolving attacks. Recent works have demonstrated that reinforcement learning can be used to train stronger attacker and defender models in tandem by applying PPO-style self-play and DPO-style online preference optimization. In this work, we explore the efficacy of GRPO in this setting. Co-training can be challenging because it requires jointly optimizing multiple properties of both the attacker and defender. We therefore shape model outputs using multiple LLM judge-based reward channels and compute advantages with GDPO, which prevents any single channel from dominating. Our method uses a curriculum that progresses from attacker-only single-turn and multi-turn training to co-training, where attacker and defender models are updated in alternation. We show that this method produces highly effective and transferable attacks, and that co-trained defenders reach competitive safety while preserving general utility. Through a controlled ablation, we further identify which components of our training pipeline most affect the resulting balance between safety and utility. Finally, we find that GRPO tends to collapse attacker diversity over training and discuss possible ways to address this limitation.

Submission history

From: Blake Bullwinkel [view email]
[v1] Mon, 8 Jun 2026 16:21:36 UTC (809 KB)
[v2] Wed, 7 Oct 2026 22:55:53 UTC (957 KB)