We audited the bridge validator of a live AI-agent protocol. Our AI found consensus-bypass flaws that could let a single party forge consensus. Details withheld under responsible disclosure.
We found the consensus check sums weights without detecting duplicate signers — the same validator appearing multiple times in a signature set counts multiple times, so a single actor could potentially reach threshold and forge a cross-chain message.
Full mechanics withheld. Project owners: contact for the report.
We found any active validator can add new validators — allowing a captured or malicious validator to dilute or take over the validator set.
Removing validators has no minimum-count guard, so the set can be reduced to an unsafe size.
Consensus flaws are the most dangerous bridge bugs — they forge trust itself. AI caught them in minutes.