MJ: Multi-turn LLM Jailbreaking via Decomposed Credit Assignment

Jul 14, 2026

A new preprint introduces DC-GRPO, a turn-level credit assignment framework for multi-turn jailbreaking of large language models (LLMs). The method assigns learning signals to individual dialogue turns, enabling more effective automated red teaming. Experiments show DC-GRPO achieves over 97% attack success rate across multiple benchmarks, substantially outperforming previous state-of-the-art methods.

Why it matters: This work demonstrates a highly effective automated jailbreaking technique for multi-turn LLM interactions, highlighting a significant vulnerability in current conversational AI safety measures.

Full story at: arXiv Computation and Language ↗