List your company

Jailbreak Attack on Prefilling Attack 40 tokens

0ASR (%)Self-Reminder
-2.678815.403133.48551.5669Feb 2, 2026

Evaluation results

MethodLinks
Self-ReminderModel=Llama2-7b2026.0201.03
Self-ExamBackbone=Qwen32026.020.611.06
Self-ReminderModel=Llama2-13b2026.020.611.08
SSDBackbone=Qwen32026.020.911.32
Self-ReminderBackbone=Qwen32026.021.211.08
SD_TinyExpertBackbone=Qwen32026.021.521.12
NGSDModel=Llama2-13b2026.021.521.37
NGSDBackbone=Qwen32026.022.121.17
Self-ReminderBackbone=Llama32026.022.731.68
No DefenseBackbone=Qwen32026.022.731.15
SSDBackbone=Llama32026.023.031.66
NGSDBackbone=Llama32026.023.641.3
Self-ExamModel=Llama2-7b2026.023.641.3
ParaphraseBackbone=Llama32026.024.851.93
Self-ExamModel=Llama2-13b2026.025.151.75
ParaphraseBackbone=Qwen32026.026.971.33
ParaphraseModel=Llama2-13b2026.026.971.62
SD_TinyExpertBackbone=Llama32026.027.271.57
Self-ReminderBackbone=Vicuna2026.027.581.58
SD_DaExpertBackbone=Qwen32026.027.581.39
NGSDModel=Llama2-7b2026.027.581.77
NGSDBackbone=Vicuna2026.028.481.77
SD_DaExpertModel=Llama2-7b2026.028.481.7
ParaphraseModel=Llama2-7b2026.02101.77
SSDModel=Llama2-13b2026.0211.521.83
SafeDecodingBackbone=Qwen32026.0213.641.55
SSDModel=Llama2-7b2026.0214.852.36
No DefenseModel=Llama2-13b2026.02202.55
Self-ExamBackbone=Llama32026.0220.912.42
SD_TinyExpertModel=Llama2-7b2026.0221.213.02
SD_DaExpertModel=Llama2-13b2026.0221.822.89
SafeDecodingModel=Llama2-13b2026.0222.122.78
ParaphraseBackbone=Vicuna2026.0225.152.72
No DefenseModel=Llama2-7b2026.0225.153.23
SafeDecodingModel=Llama2-7b2026.0226.973.33
Self-ExamBackbone=Vicuna2026.0228.182.66
SSDBackbone=Vicuna2026.02302.99
SafeDecodingBackbone=Llama32026.0234.853.31
SD_DaExpertBackbone=Llama32026.0239.73.53
SD_TinyExpertModel=Llama2-13b2026.0241.212.79
No DefenseBackbone=Llama32026.0246.363.7
SD_TinyExpertBackbone=Vicuna2026.0258.184.16
SafeDecodingBackbone=Vicuna2026.0265.764.34
No DefenseBackbone=Vicuna2026.0266.974.43
SD_DaExpertBackbone=Vicuna2026.0266.974.43