We test AuditAgents against simplified reproductions of real-world smart contract exploits to validate detection capability — no cherry-picking, no hints.
Detection Rate = PASS + PARTIAL. Clean Pass Rate = PASS only. View full methodology →
| Benchmark | Vulnerability Class | Result | Date |
|---|---|---|---|
| Parity Wallet Initialization Vulnerability | Access Control / Initialization | ✓ PASS | June 2026 |
| DAO Reentrancy Vulnerability | Reentrancy | ✓ PASS | June 2026 |
| Access Control Failure (Unlimited Minting) | Access Control | ✓ PASS | June 2026 |
| Unchecked Return Value | Error Handling | ✓ PASS | June 2026 |
| Oracle Manipulation Vulnerability | Oracle / Price Feed | ⚠ PARTIAL | June 2026 |
| bZx Flash Loan Attack | Flash Loan / Oracle | Coming Soon | — |
| Poly Network Integer Overflow | Arithmetic Overflow | Coming Soon | — |
| Nomad Bridge Merkle Root Bug | Initialization / Logic | Coming Soon | — |
All benchmarks use simplified contract reproductions of the original vulnerability class — not production source code. The goal is to evaluate vulnerability pattern detection, not to reproduce exact historical incidents.