Jailbreak Defense on ReNeLLM, EquaCode, DeepInc, WordGame
4.02ASR (ReNeLLM)FreoStream
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| FreoStreamGuardrail Category=Stream, Protected LLM=DeepSeek-R12026.06 | 4.02 | 1.24 | 10.24 | 9.9 | 6.35 | |
| FreoStreamGuardrail Category=Stream, Ablation=w/o Future-Aware Reasoning, Protected LLM=DeepSeek-R12026.06 | 4.02 | 1.24 | 11.45 | 9.14 | 6.46 | |
| Qwen-Stream-0.6BGuardrail Category=Stream, Protected LLM=DeepSeek-R12026.06 | 4.25 | 1.65 | 15.06 | 13.45 | 8.6 | |
| FreoStreamGuardrail Category=Stream, Ablation=w/o Safety-Aligned Optimization, Protected LLM=DeepSeek-R12026.06 | 4.49 | 1.65 | 15.06 | 13.7 | 8.73 | |
| Qwen-Gen-0.6BInput Context=Prompt+Response, Guardrail Category=Generative, Protected LLM=DeepSeek-R12026.06 | 6.14 | 1.24 | 25.9 | 31.98 | 16.32 | |
| Qwen-Stream-8BGuardrail Category=Stream, Protected LLM=DeepSeek-R12026.06 | 10.93 | 30.03 | 10.84 | 11.93 | 15.93 | |
| YuFeng-XGuard-0.6BInput Context=Prompt+Response, Guardrail Category=Generative, Protected LLM=DeepSeek-R12026.06 | 11.35 | 8.28 | 16.87 | 10.15 | 11.66 | |
| GuardReasoner-1BGuardrail Category=Generative, Protected LLM=DeepSeek-R12026.06 | 12.29 | 0.83 | 31.33 | 9.9 | 13.59 | |
| Nemotron-4BGuardrail Category=Generative, Protected LLM=DeepSeek-R12026.06 | 22.22 | 15.06 | 13.25 | 17.77 | 17.08 | |
| YuFeng-XGuard-0.6BInput Context=Prompt Only, Guardrail Category=Generative, Protected LLM=DeepSeek-R12026.06 | 23.88 | 13.87 | 30.72 | 25.38 | 23.46 | |
| Qwen-Gen-0.6BInput Context=Prompt Only, Guardrail Category=Generative, Protected LLM=DeepSeek-R12026.06 | 82.03 | 46.38 | 27.01 | 87.31 | 60.68 |