Prompt Injection Defense on InjecAgent
0ASRDataFilter
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DataFilterBackend LLM=Llama-3.1-8B-Instruct, Attack Type=ENHANCED2026.04 | 0 | — | |
| PromptGuardBackend LLM=Llama-3.1-8B-Instruct, Attack Type=ENHANCED2026.04 | 0 | — | |
| PIGuardBackend LLM=Llama-3.1-8B-Instruct, Attack Type=ENHANCED2026.04 | 0 | — | |
| RETABackbone=Qwen3-4B, Evaluation Protocol=Static-transfer2026.06 | 0.1 | — | |
| SecAlign++Backend LLM=Llama-3.1-8B-Instruct, Attack Type=ENHANCED2026.04 | 1 | — | |
| Attn.TrackerBackend LLM=Llama-3.1-8B-Instruct, Attack Type=ENHANCED2026.04 | 1 | — | |
| DataSentinelBackend LLM=Llama-3.1-8B-Instruct, Attack Type=ENHANCED2026.04 | 2 | — | |
| PISanitizerBackend LLM=Llama-3.1-8B-Instruct, Attack Type=ENHANCED2026.04 | 3 | — | |
| MetaSecAlignBackbone=Qwen3-4B, Evaluation Protocol=Static-transfer2026.06 | 4.2 | — | |
| PromptArmorBackend LLM=Llama-3.1-8B-Instruct, Attack Type=ENHANCED2026.04 | 5 | — | |
| No DefenseBackend LLM=Llama-3.1-8B-Instruct, Attack Type=ENHANCED2026.04 | 29 | — | |
| NoneBackbone=Qwen3-4B, Evaluation Protocol=Static-transfer2026.06 | 55.4 | — |