Question Answering on Qasper (Utility, ASR)
30ASR ScoreNo Defense
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| No DefenseAttack Type=Combined, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 30 | — | |
| PromptArmorAttack Type=Combined, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 27 | — | |
| PromptGuardAttack Type=Combined, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 27 | — | |
| PIGuardAttack Type=Combined, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 24 | — | |
| No DefenseAttack Type=Direct, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 11 | — | |
| PromptArmorAttack Type=Direct, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 11 | — | |
| PromptGuardAttack Type=Direct, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 11 | — | |
| PIGuardAttack Type=Direct, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 10 | — | |
| GPT-OSS-SafeguardAttack Type=Direct, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 4 | — | |
| AgentWatcherAttack Type=Direct, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 4 | — | |
| DataSentinelAttack Type=Direct, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 1 | — | |
| DataSentinelAttack Type=Combined, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 1 | — | |
| GPT-OSS-SafeguardAttack Type=Combined, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 1 | — | |
| AgentWatcherAttack Type=Combined, Target LLM=Qwen-3-4B-Instruct-25072026.04 | 0 | — | |
| AgentWatcherAttack Type=Clean, Target LLM=Qwen-3-4B-Instruct-25072026.04 | — | 32 | |
| DataSentinelAttack Type=Clean, Target LLM=Qwen-3-4B-Instruct-25072026.04 | — | 4 | |
| GPT-OSS-SafeguardAttack Type=Clean, Target LLM=Qwen-3-4B-Instruct-25072026.04 | — | 31 | |
| No DefenseAttack Type=Clean, Target LLM=Qwen-3-4B-Instruct-25072026.04 | — | 32 | |
| PIGuardAttack Type=Clean, Target LLM=Qwen-3-4B-Instruct-25072026.04 | — | 31 | |
| PromptArmorAttack Type=Clean, Target LLM=Qwen-3-4B-Instruct-25072026.04 | — | 31 | |
| PromptGuardAttack Type=Clean, Target LLM=Qwen-3-4B-Instruct-25072026.04 | — | 30 |