Prompt Injection Defense on SEP
0ASRPromptGuard
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PromptGuardLLM=Qwen3-4B-Instruct2026.04 | 0 | — | |
| Attn.TrackerLLM=Qwen3-4B-Instruct2026.04 | 0 | — | |
| PIGuardLLM=Qwen3-4B-Instruct2026.04 | 0 | — | |
| SecAlign++LLM=Qwen3-4B-Instruct2026.04 | 1 | 66 | |
| DataFilterLLM=Qwen3-4B-Instruct2026.04 | 1 | 98 | |
| Qwen3-1.7BSetting=ACT2026.05 | 2.3 | — | |
| Qwen3-8BSetting=ACT2026.05 | 2.7 | — | |
| PISanitizerLLM=Qwen3-4B-Instruct2026.04 | 3 | 90 | |
| Gemma-4-E4B-itSetting=ACT2026.05 | 3.5 | — | |
| Qwen3-8BSetting=BCT2026.05 | 4 | — | |
| Qwen3-1.7BSetting=BCT2026.05 | 4.8 | — | |
| Gemma-4-E4B-itSetting=BCT2026.05 | 7.1 | — | |
| GPT-OSS-20BSetting=BCT2026.05 | 14.7 | — | |
| GPT-OSS-20BSetting=ACT2026.05 | 21.7 | — | |
| DataSentinelLLM=Qwen3-4B-Instruct2026.04 | 29 | — | |
| Phi-4-reasoningSetting=ACT2026.05 | 31.7 | — | |
| Phi-4-reasoningSetting=BCT2026.05 | 45.2 | — | |
| Qwen3-1.7BSetting=Baseline2026.05 | 61.6 | — | |
| PromptArmorLLM=Qwen3-4B-Instruct2026.04 | 69 | 60 | |
| No DefenseLLM=Qwen3-4B-Instruct2026.04 | 72 | 58 | |
| GPT-OSS-20BSetting=Baseline2026.05 | 73.2 | — | |
| Gemma-4-E4B-itSetting=Baseline2026.05 | 73.6 | — | |
| Qwen3-8BSetting=Baseline2026.05 | 73.9 | — | |
| Phi-4-reasoningSetting=Baseline2026.05 | 96.5 | — |