Vulnerability Detection on Agent-Vuln-Bench (AVB) (42 vulnerabilities)
40True PositivesAgent Audit
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Agent Audit2026.03 | 40 | 2 | 6 | 95.24 | 86.96 | 0.909 | 30 | 100 | 10 | |
| Bandit2026.03 | 11 | 26 | 0 | 29.7 | 100 | 0.458 | 1 | 0 | 1 | |
| Semgrep2026.03 | 10 | 32 | 0 | 23.8 | 100 | 0.385 | 0 | 0 | 1 |