List your company

Jailbreak Safety on WildJailbreak

17.3Reasoning Harmful RatioIPO
14.69632.27349.8567.427Sep 29, 2025

Evaluation results

MethodLinks
IPOBackbone=Qwen3-8B2025.0917.3——11.4
IPOBackbone=DeepSeek-R1-Distill-Llama-8B2025.0923.4——9.6
IPOBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0923.6——12.8
GRPOBackbone=DeepSeek-R1-Distill-Llama-8B2025.0936.3——33.6
STARBackbone=DeepSeek-R1-Distill-Llama-8B2025.0937.8——22.7
STARBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0943.2——29.7
GRPOBackbone=Qwen3-8B2025.0945——33.6
GRPOBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0946——37.9
SafeKeyBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0946.6——32.9
RealSafeBackbone=DeepSeek-R1-Distill-Llama-8B2025.0947.1——2
SafeKeyBackbone=DeepSeek-R1-Distill-Llama-8B2025.0949.9——33.2
RealSafeBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0954.6——2.3
SafeChainBackbone=DeepSeek-R1-Distill-Llama-8B2025.0966.7——55.6
SafeChainBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0968.3——59.3
BaseBackbone=Qwen3-8B2025.0980——64.5
BaseBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0982.3——74.8
BaseBackbone=DeepSeek-R1-Distill-Llama-8B2025.0982.4——73.9
BackTrack2025.09—35.720.8—
Base2025.09—82.473.9—
GRPO2025.09—36.333.6—
IPO2025.09—23.49.6—
STAR2025.09—37.822.7—
TARS2025.09—33.229.4—