List your company

Jailbreak Defense on GCG

0ASRPPL
-4235077Feb 14, 2024Jun 20, 2024Oct 25, 2024Mar 1, 2025Jul 6, 2025Nov 10, 2025Mar 18, 2026

Evaluation results

MethodLinks
PPLModel=Vicuna2024.0201.02
PPLModel=Llama22024.0201.06
Self-ReminderModel=Llama22024.0201
ICDModel=Llama22024.0201
SafeDecodingModel=Llama22024.0201
Self-ExaminationModel=Gemma22025.0501.08
Goal PrioritizationModel=Gemma22025.0501
IAModel=Gemma22025.0501
SAGEModel=Gemma22025.0501
Self-ExaminationModel=Qwen2.52025.0501
Goal PrioritizationModel=Qwen2.52025.0501
SAGEModel=Qwen2.52025.0501
ICDModel=Llama3.12025.0501
SAGEModel=Llama3.12025.0501
PreSafeBase Model=DS-R1-7B2026.030—
R2DBase Model=DS-R1-8B2026.030—
PreSafeBase Model=Qwen-34B2026.030—
R2DBase Model=Qwen-8B2026.030—
R2DBase Model=DS-R1-7B2026.031—
R2DBase Model=DS-R1-14B2026.031—
PreSafeBase Model=DS-R1-14B2026.031—
RetokenizationModel=Llama22024.0221
Self-ReminderModel=Qwen2.52025.0521
IAModel=Llama3.12025.0521.08
Perplexity FilterBase Model=DS-R1-14B2026.033—
PreSafeBase Model=Skywork-7B2026.033—
SafeDecodingModel=Vicuna2024.0241.12
ParaphraseModel=Llama22024.0241.06
Self-ReminderModel=Gemma22025.0541.1
Erase-And-CheckBase Model=Qwen-8B2026.034—
PreSafeBase Model=Qwen-8B2026.034—
ICDModel=Gemma22025.0561.24
ICDModel=Qwen2.52025.0561.18
Self-ExaminationModel=Llama3.12025.0561.06
Erase-And-CheckBase Model=DS-R1-7B2026.036.1—
PreSafeBase Model=DS-R1-8B2026.036.1—
Perplexity FilterBase Model=Qwen-34B2026.036.1—
Erase-And-CheckBase Model=Qwen-34B2026.036.1—
R2DBase Model=Skywork-7B2026.036.1—
IAModel=Qwen2.52025.0571.18
Erase-And-CheckBase Model=Skywork-7B2026.037.1—
Erase-And-CheckBase Model=DS-R1-14B2026.038.1—
Perplexity FilterBase Model=DS-R1-8B2026.039.1—
Erase-And-CheckBase Model=DS-R1-8B2026.039.1—
Perplexity FilterBase Model=Qwen-8B2026.039.1—
Self-ExaminationModel=Vicuna2024.02121.4
Self-ExaminationModel=Llama22024.02121.56
Self-ReminderModel=Llama3.12025.05121.24
Perplexity FilterBase Model=Skywork-7B2026.0312.1—
Smooth LLMBase Model=DS-R1-8B2026.0314.1—
Perplexity FilterBase Model=DS-R1-7B2026.0317.2—
R2DBase Model=Qwen-34B2026.0317.2—
Smooth LLMBase Model=DS-R1-14B2026.0318.2—
SafeChainBase Model=Skywork-7B2026.0319.2—
ParaphraseModel=Vicuna2024.02201.8
Smooth LLMBase Model=DS-R1-7B2026.0321.2—
No DefenseModel=Qwen2.52025.05221.88
SafeChainBase Model=DS-R1-8B2026.0323.2—
Remove Non-DictionaryBase Model=Qwen-8B2026.0325.3—
NoneBase Model=Qwen-8B2026.0326.2—
Smooth LLMBase Model=Qwen-34B2026.0326.3—
NoneBase Model=DS-R1-14B2026.0327.2—
SafeChainBase Model=DS-R1-14B2026.0327.3—
Smooth LLMBase Model=Qwen-8B2026.0327.3—
Goal PrioritizationModel=Llama3.12025.05281.36
Remove Non-DictionaryBase Model=DS-R1-14B2026.0328.3—
NoneBase Model=Qwen-34B2026.0330.3—
No DefenseModel=Llama22024.02322.48
SafeChainBase Model=DS-R1-7B2026.0332.3—
Remove Non-DictionaryBase Model=Qwen-34B2026.0332.3—
Synonym SubstitutionBase Model=DS-R1-14B2026.0333.3—
Synonym SubstitutionBase Model=Qwen-34B2026.0334.3—
SafeChainBase Model=Qwen-34B2026.0334.3—
SafeChainBase Model=Qwen-8B2026.0335.4—
NoneBase Model=DS-R1-8B2026.0337.4—
Synonym SubstitutionBase Model=Qwen-8B2026.0337.4—
Remove Non-DictionaryBase Model=DS-R1-8B2026.0339.4—
Smooth LLMBase Model=Skywork-7B2026.0339.4—
RetokenizationModel=Vicuna2024.02421.58
Self-ReminderModel=Vicuna2024.02422.76
Synonym SubstitutionBase Model=DS-R1-8B2026.0344.4—
NoneBase Model=DS-R1-7B2026.0348.5—
Remove Non-DictionaryBase Model=DS-R1-7B2026.0348.5—
Synonym SubstitutionBase Model=Skywork-7B2026.0350.5—
Synonym SubstitutionBase Model=DS-R1-7B2026.0351.5—
NoneBase Model=Skywork-7B2026.0352.5—
Remove Non-DictionaryBase Model=Skywork-7B2026.0355.6—
No DefenseModel=Llama3.12025.05603
ICDModel=Vicuna2024.02703.86
No DefenseModel=Gemma22025.05964.74
No DefenseModel=Vicuna2024.021004.7