Jailbreak Attack Defense on GCG
0ASRSTAR-1
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| STAR-1Model Scale=Qwen3-14B2026.01 | 0 | — | — | — | |
| SafeKeyModel Scale=Qwen3-14B2026.01 | 0 | — | — | — | |
| ReasoningGuardModel Scale=Qwen3-14B2026.01 | 0 | — | — | — | |
| Self-ReminderModel Scale=Qwen3-14B2026.01 | 0 | — | — | — | |
| Self-GuardModel Scale=Qwen3-14B2026.01 | 0 | — | — | — | |
| SafeKeyModel Scale=Qwen3-4B2026.01 | 0.2 | — | — | — | |
| SafeKeyModel Scale=Qwen3-8B2026.01 | 0.2 | — | — | — | |
| Alpha-steerModel Scale=Qwen3-14B2026.01 | 0.2 | — | — | — | |
| STAR-1Model Scale=Qwen3-4B2026.01 | 0.4 | — | — | — | |
| Self-ReminderModel Scale=Qwen3-8B2026.01 | 0.4 | — | — | — | |
| Self-GuardModel Scale=Qwen3-8B2026.01 | 0.4 | — | — | — | |
| Self-ReminderModel Scale=Qwen3-4B2026.01 | 0.6 | — | — | — | |
| Self-GuardModel Scale=Qwen3-4B2026.01 | 0.6 | — | — | — | |
| STAR-1Model Scale=Qwen3-8B2026.01 | 0.6 | — | — | — | |
| ReasoningGuardModel Scale=Qwen3-4B2026.01 | 0.8 | — | — | — | |
| ReasoningGuardModel Scale=Qwen3-8B2026.01 | 0.8 | — | — | — | |
| InstructModel Scale=Qwen3-14B2026.01 | 1.5 | — | — | — | |
| Alpha-steerModel Scale=Qwen3-4B2026.01 | 3.7 | — | — | — | |
| InstructModel Scale=Qwen3-8B2026.01 | 5.4 | — | — | — | |
| Alpha-steerModel Scale=Qwen3-8B2026.01 | 5.6 | — | — | — | |
| InstructModel Scale=Qwen3-4B2026.01 | 6.2 | — | — | — | |
| SafeChainModel Scale=Qwen3-4B2026.01 | 19.6 | — | — | — | |
| SafeChainModel Scale=Qwen3-14B2026.01 | 21 | — | — | — | |
| SafeChainModel Scale=Qwen3-8B2026.01 | 23.3 | — | — | — | |
| No DefenseModel=R1-Qwen-14B2025.08 | — | — | 70.5 | 79.5 | |
| No DefenseModel=R1-Qwen-32B2025.08 | — | — | 45.5 | 63.6 | |
| ParaphraseModel=R1-Qwen-14B2025.08 | — | — | 81 | 83 | |
| ParaphraseModel=R1-Qwen-32B2025.08 | — | — | 70.5 | 75 | |
| RealSafe-R1Model=R1-Qwen-14B2025.08 | — | — | 0 | 11.4 | |
| RealSafe-R1Model=R1-Qwen-32B2025.08 | — | — | 4.5 | 25 | |
| ReasoningGuardModel=R1-Qwen-14B2025.08 | — | — | 18.2 | 29.5 | |
| ReasoningGuardModel=R1-Qwen-32B2025.08 | — | — | 4.5 | 13.6 | |
| SafeKeyModel=R1-Qwen-14B2025.08 | — | — | 9.1 | 22.7 | |
| SAFEPATH-ZSModel=R1-Qwen-14B2025.08 | — | — | 18.2 | 43.2 | |
| SAFEPATH-ZSModel=R1-Qwen-32B2025.08 | — | — | 18.2 | 38.6 | |
| Self-ReminderModel=R1-Qwen-14B2025.08 | — | — | 47.7 | 52.3 | |
| Self-ReminderModel=R1-Qwen-32B2025.08 | — | — | 22.7 | 45.5 | |
| SmoothLLMModel=R1-Qwen-14B2025.08 | — | — | 50 | 70.5 | |
| SmoothLLMModel=R1-Qwen-32B2025.08 | — | — | 54.5 | 84.1 | |
| ThinkingIModel=R1-Qwen-14B2025.08 | — | — | 2.3 | 18.2 | |
| ThinkingIModel=R1-Qwen-32B2025.08 | — | — | 2.3 | 34.1 |