Find → prove → fix → verify
About Agent-generated exploits are the primary claim. Fallback is a labeled ABI safety net only. Held-out / Blind targets require multi-step privilege escalation the safety net will not perform. Benchmark tier = attributed teaching patterns (coverage, not blind). Nemotron + Tavily + Anvil.
agent session · VulnerableVault
Re-entry agent ready. Select a target and click Run agent live.
Requires Anvil at http://127.0.0.1:8545
_