List your company

Jailbreak Attack Success Rate on JailbreakBench (ASR Variants)

91.67ASR (all)HiDRA
77.453281.144184.83588.5259Jun 13, 2026

Evaluation results

MethodLinks
HiDRABackbone=Qwen2.5-3B-Instruct, projected dimensions (m)=8192, LeakyReLU slope={0.5, 0.7}2026.0691.6770—
HiDRABackbone=Gemma2-9B-IT, projected dimensions (m)=8192, LeakyReLU slope={0.5, 0.7}2026.069057—
HiDRABackbone=Qwen2.5-7B-Instruct, projected dimensions (m)=8192, LeakyReLU slope={0.5, 0.7}2026.069069.33—
Mean-AcTBackbone=Qwen2.5-3B-Instruct2026.068768—
Mean-AcTBackbone=Qwen2.5-7B-Instruct2026.068671—
Mean-AcTBackbone=Gemma2-9B-IT2026.068453—
HiDRABackbone=Llama3.2-3B-Instruct, projected dimensions (m)=8192, LeakyReLU slope={0.5, 0.7}2026.068473.33—
Mean-AcTBackbone=Llama3.2-3B-Instruct2026.068373—
HiDRABackbone=Llama3.2-1B-Instruct, projected dimensions (m)=8192, LeakyReLU slope={0.5, 0.7}2026.0680.6776—
Mean-AcTBackbone=Llama3.2-1B-Instruct2026.067872—
Base2025.10——2
RLVR2025.10——97
SFT2025.10——77