Prompt Injection Defense on Indirect Prompt Injection (IPI) Attack Suite
0Total ASRSandboxing
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SandboxingModel=Claude Haiku 4.52026.07 | 0 | 0 | 0 | |
| DUALVIEWModel=Claude Haiku 4.52026.07 | 0 | 0 | 0 | |
| SandboxingModel=Claude Sonnet 4.62026.07 | 0 | 0 | 0 | |
| DUALVIEWModel=Claude Sonnet 4.62026.07 | 0 | 0 | 0 | |
| Output guardrailModel=Claude Sonnet 4.62026.07 | 3 | 2.2 | 4.4 | |
| Input guardrailModel=Claude Sonnet 4.62026.07 | 5.6 | 4.4 | 7.8 | |
| Input guardrailModel=Claude Haiku 4.52026.07 | 7.8 | 3.9 | 15.6 | |
| Output guardrailModel=Claude Haiku 4.52026.07 | 8.9 | 8.3 | 10 | |
| Dual LLMModel=Claude Sonnet 4.62026.07 | 11.2 | 2.2 | 30 | |
| Dual LLMModel=Claude Haiku 4.52026.07 | 17.8 | 0 | 53.3 | |
| BaselineModel=Claude Sonnet 4.62026.07 | 27.8 | 27.8 | 27.8 | |
| BaselineModel=Claude Haiku 4.52026.07 | 57 | 57.2 | 56.7 |