List your company

Robustness to Jailbreak Attacks on JBB Paraphrased

16.1Harmful Reasoning RatioIPO
13.9628.40542.8557.295Sep 29, 2025

Evaluation results

MethodLinks
IPOBackbone=DS-8B2025.0916.11.4
RealSafeBackbone=DS-8B2025.0920.43.2
BaseBackbone=DS-8B2025.0969.644.6