Permission-boundary inference on AuthBench Sensitive Task Set
94TSRFull-Access
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Full-Access2026.05 | 94 | — | — | — | — | — | — | — | — | — | — | 65.8 | |
| Gemini 3.1 Pro2026.05 | 85.8 | 74.9 | 95 | 80.9 | 94.7 | 98.5 | 95 | 58.7 | 90.3 | 64.6 | 34.8 | 28.3 | |
| Golden-Perm.2026.05 | 81.7 | — | — | — | — | — | — | — | — | — | — | 0 | |
| GPT-52026.05 | 76.7 | 69.7 | 81.8 | 69.9 | 92.5 | 97 | 92.6 | 58.8 | 66.9 | 56.3 | 33.6 | 23.3 | |
| Kimi K2.52026.05 | 70 | 53.3 | 74.8 | 57.2 | 89.3 | 96.6 | 89.4 | 51 | 84.8 | 56.8 | 74.5 | 28.3 | |
| GPT-5.3-Codex2026.05 | 65.8 | 69.3 | 75.2 | 67.2 | 80.3 | 96.3 | 83.9 | 58.1 | 73.4 | 58 | 42.6 | 15.8 | |
| Qwen3-Coder2026.05 | 63.3 | 70.5 | 63.1 | 62.6 | 91.2 | 94.5 | 91.1 | 51 | 79.5 | 53.2 | 65 | 20.8 | |
| Claude Opus 4.62026.05 | 61.5 | 58.8 | 80 | 62.3 | 93.9 | 98.4 | 94 | 61.3 | 77 | 60.7 | 47 | 25.6 | |
| GPT-5.42026.05 | 61.1 | 76.8 | 77 | 73.8 | 89 | 88.8 | 88.5 | 73.2 | 71.4 | 66.5 | 21.1 | 19.4 | |
| MiniMax M2.72026.05 | 48.3 | 69.7 | 75.3 | 68.7 | 89.4 | 92.6 | 88.2 | 53.8 | 80.7 | 56.6 | 64.2 | 15.8 | |
| Qwen3.5-397B2026.05 | 42.9 | 68 | 74.1 | 67.6 | 92.3 | 97.4 | 93.3 | 53.1 | 81.9 | 56.5 | 71.9 | 17.6 |