AFX Trade - Rekt
Five signatures and one quorum. $24.15 million gone in a single withdrawal that never should have cleared.
On July 22, 2026, AFX Trade's Arbitrum bridge did exactly what it was built to do, trust its validators, and whoever controlled five of those validator keys used that trust like a master password, moving the most of the protocol's USDC reserve into a wallet the code had no reason to doubt.
No smart contract broke, no logic failed, the bridge simply asked whether enough validators agreed the withdrawal was real, and enough of them, compromised or not, said yes.
Forty-nine days after AFX proudly announced an audit from Zellic, the same audit that documented zero test coverage and acknowledgments left unfixed, met a two-hundred-second dispute window that disputed nothing.
By the time AFX Trade suspended the bridge, the money had already crossed two networks and cleared a Uniswap auction, now resting in a wallet holding all of the funds.
When your entire security model depends on validators never being compromised, what exactly was the code protecting?
Blockaid flagged it on July 22nd, and had the outline before AFX said a word.
The alert was blunt and specific: "Blockaid detected an exploit at 2026-07-22 21:30 UTC targeting AFX, a protocol on Arbitrum. The exploit was specific to a bridge that AFX operates. Approximately 24.15M USDC has been drained thus far."
No hedging, no "investigating," just a number and a target.
Steven Goldfeder responded within the hour, and he wasn't speaking for AFX. Arbitrum's co-founder drew a boundary before the narrative could blur it: "We can confirm that the transaction in question originated from a third party protocol, and the Arbitrum native bridge has not been hacked or exploited in any way."
One paragraph to clear the base layer, while a protocol on top of it was already unwinding.
AFX Trade didn't respond until 3 hours after Blockaid’s initial post. When it did, the statement added little beyond confirmation, an incident, a paused bridge, an investigation underway. No mechanism, no new figures, no explanation of where the funds had gone.
By that point, the chain had already surfaced more detail than the official response.
PeckShield filled in the missing path within the hour. The stolen USDC had moved off Arbitrum, bridged to Ethereum, and been swapped for 12,467.5 ETH, consolidated into a single address
The flow was visible end to end. Funds don't wait for disclosure, they just leave traces.
QuillAudits then collapsed the uncertainty into two lines.
First: "The same validator set added at deployment is what signed off on this $24M withdrawal."
Then, more directly: "That points to the off-chain signing system being compromised, not the contract itself."
With that, the distinction was clear, the system hadn't been bypassed, it had been operated.
Then attention snapped back to AFX's own disclosures.
Researcher Taylor Monahan resurfaced the Zellic audit that AFX had promoted in June, with a blunt assessment: The review left important gaps, including missing test coverage and unresolved acknowledgments, and that the auditors were not given a fully runnable environment.
The critique landed directly beneath AFX's June 3rd announcement of the audit, the same post that had framed it as a milestone.
Forty-nine days separated that announcement from the withdrawal. Long enough for the audit to fade, short enough that its warnings were still current.
If the audit already outlined the risks in June, why did it take a validator quorum in July to make them visible?
The Quorum That Wasn't
A bridge like AFX's relies on off-chain signers to approve withdrawals.
Once enough voting power is assembled, the contract releases funds without further questioning. The mechanism itself is the design, the failure was in who controlled the keys.
The validator set behind AFX's bridge had been in place since the contract went live on May 12, 2026, and the same validator set signed the July 22 withdrawal.
Five hot-validator signatures, representing 7,142 of 10,000 units of voting power, cleared a threshold of 6,667. That was enough to satisfy the contract exactly as written.
Bridge Contract (exploited): 0xCb3B9A3E5668AFE84DC7A864B36b845dCE062e67
Hot-Validator Signer 1: 0x00BB84aF06daC03BFe744Da13dF9D2D6fd8e77E5
Hot-Validator Signer 2: 0x27259f90D6ae500262AcE6E8428434e0c1f308F5
Hot-Validator Signer 3: 0x2e26dE22a92e41704B3eA00cc65a6CDA47b12c9e
Hot-Validator Signer 4: 0x52D4D9AD78a53a69bD089eE8f282CE0Cd0506Da7
Hot-Validator Signer 5:
0xBB472BC3962Ad02Ac660429FdBB319B5BC66DA7b
All five carry the same "AFX Trade Exploiter" tag on Arbiscan as the loot wallet and the final consolidation address.
That's not evidence the validators acted in bad faith, it's evidence the label gets applied to any address the incident touches, signer or recipient, without distinguishing between them.
Taylor Monahan flagged that exact problem herself: The validator set had been labeled as attacker wallets too.
A 200-second dispute window sat between the withdrawal request and its finalization, and it did not stop the outcome.
n practice, that suggests a system designed to detect invalid state, not compromised signers, though that's an interpretation rather than a direct statement from the record.
The broader pattern is familiar, no smart-contract bug is required when access and signing power are enough.
Zellic's audit, which AFX promoted in June, focused on the contract's logic and code quality, while Taylor Monahan later criticized it for missing test coverage and unresolved acknowledgments.
It could verify that a quorum threshold was implemented correctly.
Whether it could verify anything about the people behind that quorum is a different question, AFX's own documentation says the report covers the bridge contract scope and shouldn't be read as a full-platform audit of every component.
If the contract did exactly what it was told, and the validators did exactly what their keys authorized, whose failure was this really?
The Ordinary Getaway
The stolen funds didn’t move like loot. They moved like routine traffic: six USDC transfers through a shared router used by ordinary users, then eight UniswapX fills on Ethereum before being consolidated into a second wallet.
Nothing in the path was custom-built for the theft, and that's what made it effective, the attacker hid inside normal behavior.
The withdrawal that started it all was created, cleared its dispute window, and finalized, releasing the full amount to the loot wallet.
Withdrawal Creation Transaction (submitted by the operational address, requesting the fraudulent payout): 0x217c45c1272550e0439e53243f2987b7fb3f58b1d33c222597bbb71851b93f74
Batched Finalize Transaction (executed once the dispute window cleared, releasing $24.15M to the loot wallet): 0x50d0b3ec6c3f5fce0f10abf81540bbb508f421494aa2b3480c4a264b0436547b
Finalizer (executes an already-authorized withdrawal, distinct from the validators who sign for it): 0x5553EA7Bda594aDE7AFe91D279779a42b2B84208
Pre-Exploit Gas / Operational Address (submitted the withdrawal creation transaction): 0x32E3200D6E944cd9bD1C8C9865293B07206e7A01The transfers left the loot wallet, now labeled AFX Trade Exploiter 1, for a shared CCTP router used by ordinary transfers.
The transfers left the loot wallet, now labeled AFX Trade Exploiter 1, and routed through BridgingKit, the shared CCTP router.
The amounts, 5,895,000, 655,000, 7,500,000, 5,000,000, 5,000,000, and 100,000 USDC, summed to the stolen total exactly.
Loot Wallet ("AFX Trade Exploiter 1"): 0x2f2974fAbc54dbA33442261211c06BD20E0FEefc
Shared Bridging Router ("BridgingKit"): 0xB3FA262d0fB521cc93bE83d87b322b8A23DAf3F0
Each transfer burned USDC on Arbitrum and minted it on Ethereum to the same recipient address, no new wallet at the other end, just a new chain.
Deposit-for-Burn Transaction (5,895,000 USDC): 0xd59a78a56165da8c32a47c89bcb7bc2bb0960962b215b65dd6d00f4ead6fedc7
Deposit-for-Burn Transaction (655,000 USDC): 0xc467e4101e570a1cf86fc9781fdb4a3a7f0618e5437608ac32c3074a3c7ff849
Deposit-for-Burn Transaction (7,500,000 USDC): 0x5489629ad1a9d8b4453eacb6c7585176956b07c7dbb5ee4d6384b462e31adee7
Deposit-for-Burn Transaction (5,000,000 USDC): 0x610a6d7058236cfed3ffd761b8bc50ec670518a7a8a7a004252d080c9475d829
Deposit-for-Burn Transaction (5,000,000 USDC): 0xb69c938d12a67c8c60c53d9ce4e02c0c7e836519a1068c49282cba035320d496
Deposit-for-Burn Transaction (100,000 USDC): 0x0e5c14b52925b9a9b088d8f08923e4a4fe677685839738ebe2c0ca66fbf97f07
The USDC was swapped into ETH through eight UniswapX Dutch-auction fills, settled by a solver labeled Rizzolver against Uniswap's Dutch Order Reactor V2.
Total received: 12,467.43703738 ETH
UniswapX Fill (3,035.80133843 ETH): 0x2c92d93539ee6c5a5a64323f9d7850da220b4b04cd781bf345beba596b9b6a2d
UniswapX Fill (338.79180965 ETH): 0x1b80124b75b7afbb7aaf018f5ca20782c9a2eab007311aef71999d1700fd7bba
UniswapX Fill (1,550.3074912 ETH): 0xe505e57a1d4fe37fac6aef6e032d5a8bb8dc4e779dfc9f7491a01cb7e5d87e0d
UniswapX Fill (2,321.39295018 ETH): 0xc88574ec75f8e3b7a3866f63b84b705d513d54eb76cea8141208ca4a44eebd7a
UniswapX Fill (2,582.07624729 ETH): 0xc140e82791a624dd66bab2eec6bb5e9c0df33198fdede931982ed07fdf42ca39
UniswapX Fill (1,292.10792468 ETH): 0x114d66d43b34d385883f3bc02c48c0de9559a6ccf7766b8c92bff06ca1a92202
UniswapX Fill (1,295.04346094 ETH): 0x0b8df70bf3227a53f81566305a252c769322c005528afdb96ca94a77c926f0e2
UniswapX Fill (51.8658944 ETH): 0xc155d394ca0b3e0b62b9da74ff30f7c6506722cadba06e0566213b1dc74401fd
One transaction moved the full ETH balance to a second wallet, tagged AFX Trade Exploiter 9.
Final Consolidation Transaction (12,467.43703738 ETH moved): 0x7469189aea7e9c23e2004a33fcf117968237bc1154aaa61276d9e68ddb5a21a4
Final Consolidation Wallet ("AFX Trade Exploiter 9"): 0x627654B2782BFc57580ecd11d40869B350b6EBaC
Since then, the funds fragmented across roughly two dozen addresses carrying real value, buried inside a much larger pile of dust-level noise, the wallet now holding a few cents where 12,467 ETH once sat.
The money moved. What happens to two dozen scattered pieces that a 30% offer was never built to reach?
The Standard Offer
AFX's second statement arrived less than an hour after its first statement, deeper into the incident and no clearer about what had actually happened.
It cited SlowMist's finding that the stolen funds remained in the attacker's address, now reported to the Crypto Defense Alliance, a collaborative network that includes multiple exchanges and ecosystem partners, to flag and freeze suspect wallets before they move.
By then, AFX's head of growth had already made the offer public: Return 70%, keep 30%, call it a bounty.
The structure was familiar, and bounty offers have become a common enough playbook after crypto hacks.
The logic is simple: Recovering most of a loss is better than recovering none of it, especially when the stolen funds still sit in a traceable wallet.
So far, the bounty has gone unanswered. The wallet sat untouched for days before fragmenting, the real value scattering across roughly two dozen addresses inside a single day, a laundering move, not a response to the offer.
AFX's hack also didn't happen in isolation. One week earlier, Ostium lost $23.75 million on the same network to a manipulated price feed.
The pattern is becoming familiar, audits, alliances, and bounties all existed before the attack, and none of them stopped it.
Days later, zeroShadow, working with SEAL, said it is highly likely the exploit is linked to UNC4899, the North Korean group tracked as TraderTraitor.
They pointed to a match between the gas-funding pattern behind AFX's attacker wallet and the one used in the KelpDAO exploit months earlier.
The fragmentation itself fits the pattern, scattering real value across two dozen addresses inside a single day, buried under a much larger layer of dust, reads like professional laundering, not a hasty smash-and-grab.
That attribution remains preliminary, but if it holds, a 30% bounty starts to look almost beside the point.
If all the usual defenses were already in place, what exactly are they protecting against?
Trust broke, and nothing else needed to.
A validator quorum said yes to a withdrawal it should never have approved, and every system built on top of that yes, the bridge, the router, the swap venue, did exactly what it was designed to do.
The dispute window counted down honestly. The bridging kit moved funds the way it moves funds for anyone.
Even the bounty offer followed a now-familiar playbook in crypto hacks.
AFX didn't fail because something broke, it failed because nothing did.
The entire $24.15 million exit ran through the bridge and a public swap, pieces nobody would think twice about.
Forty-nine days after an audit warned the codebase couldn't be fully tested, the keys turned out to be the one thing nobody tested at all.
If a hack this large can look this unremarkable at every step, how would anyone know the next one is already underway?
REKT作为匿名作者的公共平台，我们对REKT上托管的观点或内容不承担任何责任。
捐赠 (ETH / ERC20): 0x3C5c2F4bCeC51a36494682f91Dbc6cA7c63B514C
声明:
REKT对我们网站上发布的或与我们的服务相关的任何内容不承担任何责任，无论是由我们网站的匿名作者，还是由 REKT发布或引起的。虽然我们为匿名作者的行为和发文设置规则，我们不控制也不对匿名作者在我们的网站或服务上发布、传输或分享的内容负责，也不对您在我们的网站或服务上可能遇到的任何冒犯性、不适当、淫秽、非法或其他令人反感的内容负责。REKT不对我们网站或服务的任何用户的线上或线下行为负责。