{
  "id": 3124,
  "url": "https://arxiv.org/abs/2607.11070v1",
  "title": "MJ: Multi-turn LLM Jailbreaking via Decomposed Credit Assignment",
  "summary": "Modern large language models (LLMs) operate in interactive multi-turn settings, making multi-turn jailbreaking a realistic threat model and an important setting for automated red teaming. A core challenge in learning multi-turn jailbreak attackers is credit assignment: different turns contribute differently to the final outcome, yet existing learning signals are often too coarse to identify their individual contributions. We propose decomposed credit GRPO (DC-GRPO), a unified turn-level credit a",
  "authors": "Junyoung Park, Namgyu Park, Sechan Lee, Yoon-Chan Jhi, Jihoon Cho, Sangdon Park",
  "category": "research",
  "topics": "safety-alignment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-07-13T04:19:37.000Z",
  "fetched_at": "2026-07-14T16:11:46.979Z",
  "source_slug": "x-arxiv-red-teaming-query",
  "source_name": "arXiv red teaming query",
  "source_homepage": "https://arxiv.org/a/redteam",
  "ethics_ai_record_url": "https://ethics.ai/record/3124",
  "original_url": "https://arxiv.org/abs/2607.11070v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}