We audited the lending pool of a live AI-agent protocol. Our AI found a critical bad-debt risk — one that can leave a lending protocol insolvent when collateral value drops. Details are under responsible disclosure.
We found that under collateral-value drops, the liquidation path can leave the protocol with unbacked debt. Liquidators may be disincentivized from closing underwater positions, allowing bad debt to accumulate.
Full exploit mechanics withheld under responsible disclosure. Project owners: contact for the report.
The price feed is not validated for zero or stale values — which can cause wrongful liquidations or let undercollateralized positions pass.
A borrow path updates balances before performing the health check — a checks-effects-interactions violation.
Real lending-protocol flaws found by AI in minutes — including a critical issue that could make a protocol insolvent.