The Verus Bridge Got Its Money Back in May. A New Attacker Just Used the Same Hole

What Happened

At about 21:30 UTC on Wednesday, July 22, security firm Blockaid flagged an exploit draining roughly 24.15 million USDC (a dollar-pegged stablecoin) from AFX Trade, a decentralized perpetuals exchange on Arbitrum. By early Thursday, 2 more protocols had joined it. The bridge connecting the Verus blockchain to Ethereum was drained of approximately $7.

54 million in ETH, tBTC (tokenized Bitcoin), and a spread of stablecoins including USDC, USDT, and EURC. B2 Network, a Bitcoin layer 2 built on zero-knowledge proof verification, reported a loss of about $3.86 million through an attack path that had not been publicly detailed as of this writing.

On-chain tracker Lookonchain put the combined damage at roughly $35.55 million inside a window of about 6 hours.

The Verus drain is the one with history. On May 18, the same bridge lost $11.58 million to an attacker who abused the bridge's import path to trigger payouts on the Ethereum side that were never backed by deposits on the Verus side. That incident ended in a negotiated settlement: the attacker returned roughly 4,052 ETH, worth about $8.

5 million at the time, and kept 1,350 ETH as a de facto bounty. This week, Blockaid's analysis found the new attack used the same bridge contract, the same transaction entry point, and the same class of underlying vulnerability, but appears to have been carried out by a different attacker collecting funds in a freshly created wallet.

As of this writing, neither Verus nor AFX Trade had published a full post-mortem.

A negotiated return restores the balance sheet. It does not restore the security model. The Verus bridge just paid $7.54 million to demonstrate the difference.

Same Contract, Same Entry Path, Same Bug Class

The May mechanism is well documented. According to analyses by Blockaid and security firm Halborn, the bridge's Ethereum-side contract verified several elements of an incoming cross-chain import correctly but never checked that the value claimed on the source chain matched the payout being released on Ethereum.

An attacker could therefore submit a forged import message and receive real assets for a fabricated deposit: unbacked payouts, released by a contract doing exactly what it was written to do. Blockaid classified the root cause as a gap between source-chain proof verification and destination-chain value binding, the same vulnerability class behind the 2022 Wormhole and Nomad bridge hacks.

Halborn noted the May attack cost the attacker about $10 in gas and could be mitigated with roughly 10 lines of code.

What remediation Verus actually deployed between May and July has not been published as of this writing, and that gap in the public record matters less than the documented outcome: per Blockaid, the same entry path paid out again. There is also a compounding factor this series has flagged before.

The May exploit was analyzed publicly and in detail, which means the playbook was sitting in the open for 2 months. A published post-mortem of an incompletely fixed bug is not just a disclosure. It is a product specification for the next attacker, and the next attacker appears to have been someone new.

A Settlement Is Not a Patch

By industry standards, the May episode read as a success story. Most of the funds were recovered, the attacker took a bounty-sized cut, the bridge kept operating, and the incident dropped out of the news cycle. That resolution restored the balance sheet. It did not, on the evidence of this week, restore the security posture.

The reserves the settlement rebuilt are the reserves that were available to drain again, and the appearance of resolution is precisely what makes a recovered protocol feel safer than an unexploited one. It is the trust inversion this series keeps documenting, in a new location: not a verifier or a safety layer this time, but the incident-response process itself becoming the thing that misleads.

The pattern context is now dense. Supra's signature verifier accepted a zeroed signature (see 'Supra's Verifier Accepted a Zeroed Signature and Bonzo Lend Lost $9 Million'). Across took its first attack through its safety layer (see 'Across Went 5 Clean Years Without an Attack. The First One Hit Its Safety Layer').

Wanchain's treasury check signed away its own treasury 2 days ago. The fair counter-argument deserves stating: negotiated returns are rational, the May settlement recovered $8.5 million that might otherwise be gone forever, and a settlement does not preclude a fix, since teams can and often do pursue both.

All true. But from the user's side of the contract, the lesson stands: a recovered balance tells you the last attacker took a deal. It tells you nothing about whether the next one will have to.

The Key That Signs Is the Key That Steals

The AFX Trade drain, the largest of the 3, followed a different and older script. According to reporting confirmed by Blockaid, the attacker compromised the validator signing keys behind a bridge the protocol operates on Arbitrum: hot keys held off-chain by the bridge's operators.

With those keys, no vulnerability in the contract logic was needed, because the contract did what validly signed messages told it to do. Security firm PeckShield tracked the stolen USDC as it moved to Ethereum and was swapped into roughly 12,467 ETH. Offchain Labs co-founder Steven Goldfeder confirmed that Arbitrum's own native bridge was not involved, which is accurate and also clarifying: the failure lived entirely in the protocol's key management.

When a bridge's security reduces to a set of hot keys, key control is custody, and the 24.15 million USDC belonged, functionally, to whoever controlled the signers.

What This Means for You

The recurring lesson of this series is that a bridged position is a custodial position, but this week adds a sharper corollary: a protocol's incident history is not a list of closed tickets. A prior exploit that ended in a friendly settlement does not upgrade a protocol to battle-tested.

Treat any protocol that has taken 1 exploit in a given bug class as exposed to that class until you can point to a published post-mortem and an independent re-audit of the fix. If all you can find is a settlement announcement and a resumed front end, assume the bug is live, because as of this week that assumption would have been correct at least once.

The practical moves: inventory your bridged and wrapped exposure, and for every protocol on the list, check whether it has a prior incident and whether a full technical post-mortem followed. Keep long-term holdings in native assets under your own private keys. Bitcoin held in self-custody on a hardware wallet has no import path, no validator hot keys, and no settlement process to mistake for a fix.

The custodial risk that repriced $35.55 million in 6 hours this week simply does not exist in that configuration.

If you hold assets on B2 Network or AFX Trade, watch official project channels for post-mortems and any claims process, and do not interact with unsolicited recovery offers or links. Scam operations reliably follow incidents like these, targeting exactly the users most anxious to get funds back.

What to Watch

Watch whether Verus publishes a full post-mortem this time, and specifically whether it accounts for what was and was not fixed between May and July, which is the single most informative document this incident could produce. Watch the AFX attacker's roughly 12,467 ETH: the settlement playbook cuts both ways, and either a negotiated return or a laundering path will show up on-chain first.

Watch whether B2 Network details its attack path, which as of this writing remains undisclosed. And watch whether the industry's incident-response norm shifts from negotiated returns toward independent re-audits as a condition of bridges reopening. No evidence of coordination between the 3 attacks has been published as of this writing, and 3 unrelated drains in 6 hours is, if anything, the more uncomfortable reading.

Tracked entities this report names, with total coverage across the archive. Follow one into the Actor Network to see every report it appears in.

Other reports that name the same tracked actors — a different thread than “related,” which follows wording. These are linked by who appears in them.