List your company

Jailbreak Defense on AIM AdvE

0Attack Success Rate (ASR)JPU
-0.43162.48175.3958.3083Jan 6, 2026

Evaluation results

MethodLinks
JPUBase Model Architecture=Llama3-8B-Instruct2026.010
JPUBase Model Architecture=Llama2-7B-Chat2026.010.01
RSFTBase Model Architecture=Llama3-8B-Instruct2026.010.24
CKUBase Model Architecture=Llama3-8B-Instruct2026.010.24
RSFTBase Model Architecture=Llama2-7B-Chat2026.010.48
CKUBase Model Architecture=Llama2-7B-Chat2026.010.48
Safe UnlearningBase Model Architecture=Llama2-7B-Chat2026.010.72
Circuit BreakerBase Model Architecture=Llama2-7B-Chat2026.010.72
Safe UnlearningBase Model Architecture=Llama3-8B-Instruct2026.010.72
Circuit BreakerBase Model Architecture=Llama3-8B-Instruct2026.010.72
EraserBase Model Architecture=Llama3-8B-Instruct2026.016.95
EraserBase Model Architecture=Llama2-7B-Chat2026.018.15
Base modelBase Model Architecture=Llama3-8B-Instruct2026.019.83
Base modelBase Model Architecture=Llama2-7B-Chat2026.0110.79