Integrity Protection on AgentDojo Integrity
97AUROCGIF
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GIFModel=QWEN3-14B, PSP@k Cutoff=@502026.06 | 97 | 89 | |
| GIFModel=QWEN3-32B, PSP@k Cutoff=@502026.06 | 96 | 81 | |
| GIFModel=QWEN3-8B, PSP@k Cutoff=@502026.06 | 93 | 72 | |
| GIFModel=GPT-OSS-20B, PSP@k Cutoff=@502026.06 | 92 | 65 | |
| GIFModel=Average, PSP@k Cutoff=@502026.06 | 91 | 70 | |
| GIFModel=GPT-OSS-120B, PSP@k Cutoff=@302026.06 | 88 | 77 | |
| GIFModel=GPT-OSS-120B, PSP@k Cutoff=@502026.06 | 88 | 73 | |
| GIF-Model=GPT-OSS-120B, PSP@k Cutoff=@502026.06 | 88 | 76 | |
| GIF-Model=GPT-OSS-120B, PSP@k Cutoff=@302026.06 | 87 | 75 | |
| GIFModel=GPT-OSS-120B, PSP@k Cutoff=@152026.06 | 85 | 74 | |
| GIFModel=GPT-OSS-120B, PSP@k Cutoff=@102026.06 | 83 | 69 | |
| GIF-Model=GPT-OSS-120B, PSP@k Cutoff=@152026.06 | 83 | 70 | |
| GIF-Model=GPT-OSS-120B, PSP@k Cutoff=@102026.06 | 81 | 66 | |
| GIFModel=GEMMA-4-31B, PSP@k Cutoff=@502026.06 | 80 | 38 | |
| GIFModel=GPT-OSS-120B, PSP@k Cutoff=@52026.06 | 79 | 62 | |
| GIF-Model=GPT-OSS-120B, PSP@k Cutoff=@52026.06 | 77 | 61 | |
| GIFModel=GPT-OSS-20B, PSP@k Cutoff=@52026.06 | 77 | 45 | |
| GIFModel=Average, PSP@k Cutoff=@52026.06 | 75 | 49 | |
| RTBASModel=GPT-OSS-120B, PSP@k Cutoff=@502026.06 | 69 | 35 | |
| RTBASModel=GPT-OSS-120B, PSP@k Cutoff=@302026.06 | 62 | 32 | |
| RTBASModel=GPT-OSS-120B, PSP@k Cutoff=@152026.06 | 52 | 25 | |
| RTBASModel=GPT-OSS-120B, PSP@k Cutoff=@52026.06 | 51 | 24 | |
| RTBASModel=GPT-OSS-120B, PSP@k Cutoff=@102026.06 | 51 | 24 |