ShallowDefend: Harnessing Shallow Safety Alignment for Multi-Turn Defender

Summary

The paper addresses multi-turn jailbreaks, where a harmful request spread across a conversation still succeeds despite safety alignment that concentrates in a model's first few output tokens. It proposes Assess → Decide → Respond, which has the defender commit to an explicit, parseable safety decision before generating a reply, trained with reinforcement learning at the payload turn. Across two model families and fourteen benchmarks, the authors report the strongest multi-turn safety in a controlled comparison, with a modest cost to helpfulness.

Related items

• InfoHow I Aligned Your Model: Scaling Up Multilingual Safety Training with TemplatesSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoRed-Teaming Large Language Models in Low-Resource African Languages: A Case Study of Yoruba and Nigerian PidginSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoSystematic Evaluation of Adversarial Jailbreaking Attacks and Defenses Across Seven Large Language ModelsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoDiscovering Jailbreaks via Curriculum of ModelsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoAdversarial Self-DistillationSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)