Jailbreak Detection on XGuard (train)
0.908AUCSAFEDREAM
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SAFEDREAMMethod category=Multi-turn stateful detectors (external), Target LLM=Qwen2.5-7B-Instruct2026.04 | 0.908 | 89.1 | 7.9 | 1.11 | |
| TARSMethod category=Alignment-based methods (modify LLM weights), Target LLM=Qwen2.5-7B-Instruct2026.04 | 0.895 | 87.6 | 5.9 | 0.38 | |
| X-BoundaryMethod category=Alignment-based methods (modify LLM weights), Target LLM=Qwen2.5-7B-Instruct2026.04 | 0.877 | 86.2 | 7 | 0.33 | |
| Circuit BreakerMethod category=Alignment-based methods (modify LLM weights), Target LLM=Qwen2.5-7B-Instruct2026.04 | 0.845 | 83.4 | 9.4 | 0.25 | |
| SelfDefendMethod category=external, stateless, Target LLM=Qwen2.5-7B-Instruct2026.04 | 0.818 | 61.6 | 22 | 0.07 | |
| BingoGuardMethod category=external, stateless, Target LLM=Qwen2.5-7B-Instruct2026.04 | 0.752 | 63.9 | 17.6 | 0.45 | |
| GuardReasonerMethod category=external, stateless, Target LLM=Qwen2.5-7B-Instruct2026.04 | 0.726 | 59.7 | 14.7 | 0.18 | |
| R2-GuardMethod category=external, stateless, Target LLM=Qwen2.5-7B-Instruct2026.04 | 0.688 | 64.8 | 18.8 | 0.59 | |
| WildGuardMethod category=external, stateless, Target LLM=Qwen2.5-7B-Instruct2026.04 | 0.65 | 71.3 | 22.4 | 0.53 |