List your company

Jailbreak Attack on Harmful instructions

0ASR (Harmful Instructions)Vanilla
-3.821.8547.573.15Mar 8, 2025

Evaluation results

MethodLinks
VanillaTarget Model=Llama3.2 1b2025.030
nanoGCGTarget Model=Llama3.2 3b2025.030
VanillaTarget Model=Llama3.2 3b2025.030
VanillaTarget Model=Gemma2 2b2025.030
nanoGCGTarget Model=Llama3.2 1b2025.036
VanillaTarget Model=Qwen2.5 1.5b2025.0312
Attention-SSRTarget Model=Llama3.2 1b2025.0314
Attention-SSRTarget Model=Llama3.2 3b2025.0328
Attention-SSRTarget Model=Gemma2 2b2025.0334
Steering-SSRTarget Model=Llama3.2 3b2025.0340
Steering-SSRTarget Model=Llama3.2 1b2025.0353
nanoGCGTarget Model=Gemma2 2b2025.0353
Steering-SSRTarget Model=Gemma2 2b2025.0354
Probe-SSRTarget Model=Llama3.2 3b2025.0380
Probe-SSRTarget Model=Gemma2 2b2025.0384
nanoGCGTarget Model=Qwen2.5 1.5b2025.0385
Probe-SSRTarget Model=Llama3.2 1b2025.0388
Probe-SSRTarget Model=Qwen2.5 1.5b2025.0391
Attention-SSRTarget Model=Qwen2.5 1.5b2025.0394
Steering-SSRTarget Model=Qwen2.5 1.5b2025.0395