Two files, eleven findings, five assumptions

OpenZeppelin reviewed the Bridge2 contract deployed on Arbitrum One, together with Signature.sol — a custodial USDC bridge between Arbitrum and TxFlow L1. The engagement ran from 25 to 30 June 2026 and covered two Solidity files. The report is dated 19 August.

Eleven issues: none critical, none high, one medium, five low, five notes. Five were resolved outright and one partially; the notes were deferred to future updates.

The medium finding

updateValidatorSet had no replay protection. A signed update could be resubmitted, and each resubmission reset the dispute period — which means finalisation could be pushed out indefinitely without forging anything. The recommendation was to call checkMessageNotUsed on the message. It is a denial-of-finality bug rather than a theft bug, and those tend to be underweighted precisely because nothing moves.

The low findings worth reading

• Malicious validator set parameters could permanently disable the bridge through epoch overflow, power overflow or an excessive validator count.
• No lower bound on locker threshold updates — later documented as intentional.
• A non-standard EIP-712 implementation, so users signing did not see what they were signing in the usual form.
• Deposits accepted while the bridge was paused.
• Permit front-running on deposits.

The part that is not a bug

The report sets out what the design assumes rather than proves: that no more than a third of validator power is compromised; that lockers can pause within the dispute window before finalisation; that hot and cold wallet separation permits recovery from a hot key compromise; that the L1 manages the validator set and role lifecycle correctly; and that the Arbitrum sequencer stays honest.

One low-severity item — the validator count overflow — was closed as an L1 trust assumption rather than an on-chain code responsibility. That is a defensible call and also a good illustration of what a two-file scope does. The contracts were found to be sound against their assumptions. Whether the assumptions hold is a question the audit was not asked, and anyone sizing exposure to this bridge should read the assumption list as the actual risk register.