Security Analysis on Security Tasks 15,000 benign samples (test)
96.8F1 (Log)SecureCAI
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SecureCAI2026.01 | 96.8 | 95.4 | 93.2 | 95.1 | |
| Base LLM2026.01 | 94.2 | 91.8 | 88.6 | 91.5 | |
| Standard CAI2026.01 | 92.7 | 90.1 | 86.4 | 89.7 | |
| Instr. Hierarchy2026.01 | 89.4 | 87.2 | 83.5 | 86.7 | |
| Input Filtering2026.01 | 78.3 | 75.6 | 71.2 | 75 |