List your company

Jailbreak Defense on HarmBench and AdvBench (test)

91.2GCG ScoreNo Defense
-2.08822.13146.3570.569Jan 15, 2026

Evaluation results

MethodLinks
No DefenseVictim Model=Vicuna-7B2026.0191.285.379.594.841.295.6
No DefenseVictim Model=Qwen2.5-7B2026.0185.280.475.192.438.694.5
No DefenseVictim Model=Mistral3-8B2026.0184.375.773.589.236.490.5
Self-reminderVictim Model=Vicuna-7B2026.0182.676.370.216.212.411.5
Self-reminderVictim Model=Qwen2.5-7B2026.0179.266.845.312.48.59.6
No DefenseVictim Model=Llama3-8B2026.0178.468.264.582.132.484.6
Self-reminderVictim Model=Mistral3-8B2026.0178.364.742.615.410.29.8
Self-reminderVictim Model=Llama3-8B2026.0172.559.438.210.57.28.4
R2D2Victim Model=Vicuna-7B2026.0135.541.633.221.415.817.6
R2D2Victim Model=Qwen2.5-7B2026.0132.438.635.218.412.615.2
MARTVictim Model=Vicuna-7B2026.0129.316.619.49.212.27.1
R2D2Victim Model=Mistral3-8B2026.0128.633.430.216.511.213.6
MARTVictim Model=Qwen2.5-7B2026.0126.812.113.48.510.46.2
CATVictim Model=Qwen2.5-7B2026.0126.231.428.614.89.511.4
MARTVictim Model=Mistral3-8B2026.0125.311.412.38.810.26.5
CATVictim Model=Vicuna-7B2026.0124.332.826.317.212.414.6
CATVictim Model=Mistral3-8B2026.0122.528.124.713.49.611.8
MARTVictim Model=Llama3-8B2026.0122.410.211.57.28.55.1
PPLVictim Model=Vicuna-7B2026.0119.181.573.186.435.221.4
SmoothLLMVictim Model=Qwen2.5-7B2026.0118.431.229.510.26.45.2
SmoothLLMVictim Model=Mistral3-8B2026.0116.229.727.412.87.26.1
SmoothLLMVictim Model=Llama3-8B2026.0115.228.525.48.55.44.1
PPLVictim Model=Qwen2.5-7B2026.0114.568.458.285.632.418.2
R2D2Victim Model=Llama3-8B2026.0114.412.610.29.87.58.4
SmoothLLMVictim Model=Vicuna-7B2026.0114.331.22813.58.16.8
SafeDecodingVictim Model=Qwen2.5-7B2026.0113.416.518.95.13.21.8
CATVictim Model=Llama3-8B2026.0113.113.711.28.66.87.9
SafeDecodingVictim Model=Mistral3-8B2026.0112.415.217.75.63.32.1
SafeDecodingVictim Model=Vicuna-7B2026.0112.212.610.46.84.12.6
CircuitBreakersVictim Model=Vicuna-7B2026.0111.613.413.25.43.52.2
SafeDecodingVictim Model=Llama3-8B2026.0111.213.415.24.12.82.4
PPLVictim Model=Mistral3-8B2026.0111.267.756.781.230.514.8
CircuitBreakersVictim Model=Qwen2.5-7B2026.0110.213.416.24.22.51.5
PPLVictim Model=Llama3-8B2026.0110.258.449.275.828.512.5
CircuitBreakersVictim Model=Mistral3-8B2026.019.312.615.44.92.93.7
SSPVictim Model=Vicuna-7B2026.018.86.77.42.63.45.1
ACE-safetyVictim Model=Vicuna-7B2026.018.59.87.36.15.44.2
CircuitBreakersVictim Model=Llama3-8B2026.018.511.212.43.42.13.2
SSPVictim Model=Mistral3-8B2026.018.56.57.32.83.72.7
ACE-safetyVictim Model=Mistral3-8B2026.018.19.19.57.46.85.1
ACE-safetyVictim Model=Llama3-8B2026.014.53.84.25.24.53.2
ACE-safetyVictim Model=Qwen2.5-7B2026.012.53.12.95.24.13.1
SSPVictim Model=Qwen2.5-7B2026.011.72.41.41.32.13
SSPVictim Model=Llama3-8B2026.011.52.21.31.52.43.5