Jailbreak Robustness on WJ (heldout)
5Attack Success Rate (ASR)BCT
Evaluation Results
| Method | Links | |
|---|---|---|
| BCTModel=Qwen3-4B-Instruct-2507, Stage=Pre2026.06 | 5 | |
| MLPCTModel=Qwen3-4B-Instruct-2507, Stage=Pre2026.06 | 5 | |
| AttCTModel=Qwen3-4B-Instruct-2507, Stage=Pre2026.06 | 5 | |
| BCTModel=Qwen3-4B-Instruct-2507, Stage=Post2026.06 | 6 | |
| MLPCTModel=Qwen3-4B-Instruct-2507, Stage=Post2026.06 | 6 | |
| BCTModel=Qwen3-8B, Stage=Pre2026.06 | 17 | |
| MLPCTModel=Qwen3-8B, Stage=Pre2026.06 | 17 | |
| AttCTModel=Qwen3-8B, Stage=Pre2026.06 | 17 | |
| MLPCTModel=Llama-3.1-8B-Instruct, Stage=Post2026.06 | 20 | |
| AttCTModel=Gemma-3-4B-IT, Stage=Post2026.06 | 21 | |
| BCTModel=Llama-3.1-8B-Instruct, Stage=Post2026.06 | 22 | |
| BCTModel=Qwen3-8B, Stage=Post2026.06 | 22 | |
| AttCTModel=Qwen3-4B-Instruct-2507, Stage=Post2026.06 | 25 | |
| AttCTModel=Qwen3-8B, Stage=Post2026.06 | 25 | |
| AttCTModel=Gemma-3-27B-IT, Stage=Post2026.06 | 27 | |
| AttCTModel=Llama-3.1-8B-Instruct, Stage=Post2026.06 | 27 | |
| MLPCTModel=Qwen3-8B, Stage=Post2026.06 | 27 | |
| BCTModel=Gemma-3-27B-IT, Stage=Pre2026.06 | 29 | |
| MLPCTModel=Gemma-3-27B-IT, Stage=Pre2026.06 | 29 | |
| AttCTModel=Gemma-3-27B-IT, Stage=Pre2026.06 | 29 | |
| BCTModel=Llama-3.1-8B-Instruct, Stage=Pre2026.06 | 31 | |
| MLPCTModel=Llama-3.1-8B-Instruct, Stage=Pre2026.06 | 31 | |
| AttCTModel=Llama-3.1-8B-Instruct, Stage=Pre2026.06 | 31 | |
| BCTModel=Gemma-3-4B-IT, Stage=Pre2026.06 | 32 | |
| MLPCTModel=Gemma-3-4B-IT, Stage=Pre2026.06 | 32 | |
| AttCTModel=Gemma-3-4B-IT, Stage=Pre2026.06 | 32 | |
| MLPCTModel=Gemma-3-4B-IT, Stage=Post2026.06 | 34 | |
| BCTModel=Gemma-3-27B-IT, Stage=Post2026.06 | 37 | |
| BCTModel=Gemma-3-4B-IT, Stage=Post2026.06 | 38 | |
| MLPCTModel=Gemma-3-27B-IT, Stage=Post2026.06 | 38 |