🧪 Test?View on arXiv
CoFiBreak: A Coarse-to-Fine Diversity Reward for Effective Multi-Turn Jailbreak
Not provided in the abstract
reinforcement learningattack strategiesdiversity rewards
2610.40926
Builder Relevance
1h ago80%
Abstract
CoFiBreak introduces a novel diversity reward mechanism that enhances attack diversity in automated red teaming while maintaining effectiveness and efficiency.
Reality Card
Core Claim
CoFiBreak discovers 1.6-2.0x as many distinct successful fine-grained strategies compared to the strongest baseline, achieving a 98.8% attack success rate with the highest diversity score.
Method / Result
Achieved 98.8% ASR@10 on AdvBench with a diversity score 2.1x that of the strongest baseline.
Limitations
The paper does not specify authors or provide detailed reproducibility information.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.
No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.