Metraipolis › archive › dc_d6c985be5dea

CoFiBreak: A Coarse-to-Fine Diversity Reward for Effective Multi-Turn Jailbreak — arXiv:2610.40926

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.40926
License (as archived)
CC0-1.0
Archived
2026-11-03T07:54:32.981Z
SHA-256 of the archived text
53e19db0dbf088c3e1c821abb7f9d63f0e7515f395c4977b0d3462de08de1f61
Stable URL of this page
https://metraipolis.com/documento/dc_d6c985be5dea

How to cite

CoFiBreak: A Coarse-to-Fine Diversity Reward for Effective Multi-Turn Jailbreak — arXiv:2610.40926 — arXiv (en), https://arxiv.org/abs/2610.40926 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_d6c985be5dea (Metraipolis archive dc_d6c985be5dea, sha256 53e19db0dbf088c3e1c821abb7f9d63f0e7515f395c4977b0d3462de08de1f61)
BibTeX
@misc{dcd6c985be5dea,
  title = {CoFiBreak: A Coarse-to-Fine Diversity Reward for Effective Multi-Turn Jailbreak — arXiv:2610.40926},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.40926},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_d6c985be5dea; sha256 53e19db0dbf088c3e1c821abb7f9d63f0e7515f395c4977b0d3462de08de1f61},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.40926. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Automated red teaming trains an attacker large language model (LLM) to elicit harmful responses from a victim LLM, but reinforcement learning can collapse attackers onto a narrow set of prompts when attack success is the only reward. Such a collapse of attack diversity can bias safety evaluation toward repeatedly testing near-duplicate attack patterns rather than the underlying harmful intents. Existing methods often trade off attack success or inference cost for greater content or strategy diversity. We propose CoFiBreak, a coarse-to-fine diversity reward for Group Relative Policy Optimization (GRPO) that improves attack diversity while preserving strong attack effectiveness and low inference cost. The coarse reward broadens which strategies the attacker uses by encouraging strategies underrepresented within each GRPO group, while the fine reward broadens how those strategies are realized by discouraging reuse of the same personas, framings, and output formats after harmful-intent information is removed from prompt embeddings. Diversity rewards apply only to attacks that already exceed an effectiveness threshold, so diversity cannot be gained through ineffective attacks. Since both rewards are computed only during training, CoFiBreak also retains inference efficiency. Trained with Llama3.1-8B-Instruct as the victim, CoFiBreak discovers 1.6-2.0x as many distinct successful fine-grained strategies as the strongest baseline across frontier victims on HarmBench. It reaches 98.8% ASR@10 (attack success rate with 10 generations per intent) on AdvBench with the highest diversity score among all compared methods, which is 2.1x that of the strongest-attacking baseline. The resulting attacks transfer to unseen frontier LLMs, reaching 71.0% ASR@10 on AdvBench with the unseen victim Claude-Opus-4.6. The code will be released.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.40926 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-03T07:49:45.983Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).