K3 Capital sBOLD hack: a copycat bot and block builders kept 59% of the $3.42M
Six wallets raced to empty a Liquity savings vault on 18 November. We read all 97 of their transactions to see who ended up with the money.
- $3.42M
- 59%
- 56%
- 0
- 12:14
- 162 ETH
01 · What happenedK3 Capital's sBOLD vault lost $3.42M in eight minutes
On 18 November, a savings vault on Ethereum lost everything it held. It took eight minutes.
The vault is sBOLD, built by K3 Capital. People put BOLD in it, a dollar stablecoin, and earn interest. Just before the attack, it held about $3.42M worth of BOLD. By the end, almost nothing was left.
Blockaid, a security firm, raised the alarm on X, and its post put the loss at about $3.43 million across about 34 transactions. K3 Capital said soon after that it was looking into the attack and had paused the vault. Crypto news sites repeated the alert.
Bitquery reads every transaction on Ethereum, so we went through the drain block by block. Blockaid had the total right. But 97 transactions emptied the vault, and six wallets sent them. The wallet that first put the attack on chain, along with a second wallet it funded, ended up with about $0.52M. A copycat bot took 56%, and it handed most of that to the companies that build Ethereum's blocks.
We also checked how the attack worked. The first alerts said the attackers liquidated their own loans to get ETH into the vault. No loan was liquidated.
02 · The vaultWhat sBOLD is: a savings vault for Liquity's BOLD
BOLD is the stablecoin of Liquity V2, a lending protocol on Ethereum. People mint BOLD by borrowing against ETH, wstETH or rETH. The last two are tokens that earn staking rewards. Each of these loans is called a trove.
Liquity also runs three Stability Pools, one for each type of collateral that borrowers lock up. People deposit BOLD there to earn interest and to back the loans. When a loan's collateral falls too low, the pool pays off the loan with its BOLD and takes the collateral in return.
K3 Capital built sBOLD on top of those pools. You deposit BOLD and get sBOLD shares back, and the vault spreads the BOLD across all three pools. Liquity calls it an on-chain savings account. A share's price follows how much BOLD the vault holds, so it rises as interest comes in.
When the pools hand the vault collateral, such as WETH (the token form of ETH), the vault has to turn it back into BOLD. It does that through a function called swap(). Anyone can call it, and the vault can pay the caller a reward.
03 · The flawA flaw ChainSecurity flagged last year
ChainSecurity, an audit firm, checked sBOLD's code for K3 Capital. Its report rated several issues high, and one of them matters here. When the vault priced its shares, it ignored any collateral it held below a set amount, which the auditor's example put at 7,500 BOLD.
That gap lets someone buy shares cheap and sell them for more. You get some ETH into the vault, so it holds more than its share price shows. You deposit while that ETH is not counted. You call swap(), which turns it into BOLD that does count, and the share price jumps. Then you cash out at the higher price, and do it again.
In the auditor's example, the ETH came in through a liquidation. The report marks the issue as fixed in a later version of the code. We did not check which version K3 Capital runs. ExVul, another security firm, wrote that the drain matches the path in this audit.
04 · One roundHow the sBOLD exploit worked, round by round
Blockaid flagged one transaction, sent by a wallet whose address starts with 0xdc0e. We call it the attacker. It opened with a flash loan, a loan that has to be paid back inside the same transaction. The attacker borrowed 10,000 wstETH from the lending protocol Morpho, used it as collateral for a trove, and minted about 9 million BOLD.
Next came the same round, eight times over. The attacker deposited the BOLD into sBOLD and sent the vault a little WETH, just under the amount in the auditor's example. It called swap(), and the vault sold that WETH for BOLD, which did count. Then it cashed out its shares for about 10,600 BOLD more than it had paid in.
To finish, the attacker closed the trove and paid back the flash loan. It handed the leftover BOLD to Liquity for WETH, wstETH and rETH, a swap that Liquity calls a redemption. That one drain cleared about $23,000.
Blockaid's alert said the attackers liquidated their own trove to push ETH into the pools, and ExVul said the same. We read every trove event on all three of Liquity's markets during the drain, using Bitquery's smart contract data. Troves were opened and closed 95 times each, and there were 301 redemptions. Not one trove was liquidated. In each round, the attacker sent in the WETH itself.
05 · The raceSix wallets raced through 97 transactions
The first drain did not come from the attacker. A wallet whose address starts with 0x2324 drained the vault first, four times in one block and four more in the next. We call it the first copier. The attacker's own first drain landed 24 seconds later.
Two more wallets, starting with 0xbabe and 0xf601, joined about a minute later. So did a MEV bot, a program that watches pending Ethereum transactions and copies any that make money, or jumps ahead of them. MEV stands for maximal extractable value, the profit from choosing the order of transactions in a block.
From then on, the bot landed at least one drain in almost every block. Eight minutes after the first drain, almost nothing was left, and the race was over. K3 Capital's post about the pause came 35 minutes later.
Blockaid counted about 34 transactions, close to what the wallets ExVul named had sent. The whole drain took 97 transactions, and the bot sent more than half of them.
06 · The copyWho wrote the attack, and who copied it
ExVul named the first copier as the attacker. Its code says otherwise. It is a copy of the attacker's contract.
The attacker's wallet got its first ETH less than an hour before the drain. Then, 11 minutes before the first drain, it deployed a contract and never used it. The first copier's two drain contracts are the same 9,132 bytes, with two addresses swapped. One of the two is the deployer's own address. Where the attacker's code names the attacker's wallet, the copy names the copier's.
The first copier is not a fresh wallet. It has sent 978 transactions since late 2025, and our labels mark it as a MEV bot. Our reading is that it took the attacker's code as soon as it went on chain, and ran it first. Bots that copy other people's profitable code like this are called generalized frontrunners.
The attacker then sent 1 ETH to a second wallet, starting with 0x8fb8. That wallet put the same code on chain under its own address and drained the vault in three of the last four blocks.
07 · The botA MEV bot took 56% of the money
The copycat bot sent its transactions from 13 wallets whose addresses start with a long run of zeros. All of them called one bot contract.
Its transactions pulled 56% of the money out of the vault, more than the five early wallets took between them. They redeemed the BOLD with Liquity and swapped the proceeds to ETH on the spot. As the drain went on, each one drew more BOLD, about twice as much near the end as at the start. Bots like this race each other all over DeFi, and the bots that clear DeFi's bad loans do it every day.
08 · The bribesBlock builders got $1.78M in bribes
Ethereum blocks are put together by block builders, companies that choose which transactions go in and in what order. A bot that needs to be first pays the builder directly, inside its own transaction. Traders call that payment a bribe, or a tip.
The bot paid most of its take to builders this way, about nine ETH in every ten. It kept the rest in one payout wallet.
The bot paid 12.5 ETH for its first drain, and the price rose to 44 ETH for one block near the end. Titan built most of those blocks, and BuilderNet, Ultrasound and Quasar built the rest. Paying a builder for a place in the block is normal practice in Ethereum MEV. We have no sign that the builders knew what the bot was doing.
The first copier also paid builders, more than half of the ETH it took. Together, the builders received $1.78M, more than half of what the vault lost. Builders have done well out of a hack before. When the Binance hacker cashed out, builders and validators took $30 million.
09 · The exitsWhere the money went: THORChain, Maya and fresh wallets
The attacker and its second wallet took about $0.52M between them. Within three hours of the drain, almost none of it was left in their wallets.
The attacker sent 162 ETH to three THORChain vaults. THORChain is a network for swapping coins across chains. A deposit usually carries a memo that says which coin to buy and where to send it. These three carried none, and the chain does not say where the ETH was meant to go.
From early afternoon, both wallets also sent ETH into Maya Protocol, a fork of THORChain. Each deposit's memo asked for Bitcoin, at five different addresses. Maya sent part of two orders back, and the attacker sent the second refund straight back in. Net of the refunds, 129 ETH went in. By 15:30 UTC, two of the Bitcoin addresses had received 3.10 BTC between them. The other three had received nothing when we checked.
The other early wallets have moved less, though one of them emptied its drain contracts within two hours and sent 100 ETH to a fresh wallet. That wallet had never been used. Another, with its bot contract, still held about 95 ETH. The first copier sent most of what it kept to a single wallet, which has not moved it.
| Held at 16:14 UTC, 18 November | Balance |
|---|---|
| 0xf601 and its bot | 94.7 ETH, 3.61 rETH |
| Fresh wallet 0x97ae | 100 ETH, from 0xbabe |
| 0xbabe | 19.8 ETH |
| 0x426e | 39.4 ETH, from 0x2324 |
| 0x2324 | 2.78 ETH, 3.22 rETH |
| Bot payout wallet | 149.9 ETH |
| 0xdc0e and 0x8fb8 | 0.01 ETH |
Put together, the block builders and the copycat bot's payout wallet kept 59% of what the vault lost. The builders alone made more than three times as much from this attack as the wallet that first put it on chain.
10 · The limitsWhat the chain cannot tell us
We do not know who controls any of these wallets. Wallets can be linked by the money between them, and by shared code, but that does not name a person.
We did not read the source code K3 Capital runs. So we cannot say whether the fix in ChainSecurity's report shipped, or whether this drain used a path the fix missed.
We cannot see what THORChain will do with deposits that carry no memo. Nor can we see whether Maya paid the other Bitcoin addresses after we checked.
sBOLD is also traded on Pendle as fixed- and variable-yield tokens. People who held those lost value too. We did not measure that loss.
11 · The methodHow we traced the K3 Capital hack
We listed every transaction that minted sBOLD shares on 18 November. For each one, we read the BOLD that left Liquity's three Stability Pools and the ETH and rETH that reached the sender's own wallets. For each group, what it kept was between 88% and 95% of what it drew from the pools. The gap is fees and swap costs.
We read the vault's balance block by block from a public Ethereum node. Bribes are payments to the address that collected each block's fees. Prices come from Bitquery's price data, and service names from our own labels. We left out the fake tokens that bots sent to these wallets, a trick called address poisoning.
The full Ethereum history is also sold as files, as the Ethereum dataset on the Bitquery Data Store. For cases like this one, we also run crypto investigation services.
Ethereum balances and the vault's state were read at 16:14 UTC on 18 November 2026, at block 26,443,076. The wallets were still moving money then. Anything after that is not covered.
K3 Capital has said only that it is investigating and has paused the contracts. Its own report may describe the event differently.
The wallets are tied to each other by the transfers between them and by shared code. We attribute them to nobody.
That 0x2324 copied 0xdc0e's contract is our reading of when each contract went on chain and of the matching code. We cannot see how it got the code.
The group values add the ETH and rETH that reached each group's own wallets, at $1,801 for ETH and Bitquery's daily price for rETH. The vault's loss uses Bitquery's daily BOLD price.
The bribes are payments to the address that collected each block's fees. We do not say any builder knew what the transactions did.
THORChain and Maya amounts are what went in. On Bitcoin, we checked only what the memo addresses received, not where the coins went next.
12 · The recordThe addresses behind the sBOLD drain
Follow the money in plain English
Every transfer in this story comes from Bitquery's data. The Bitquery MCP server puts that data behind an AI assistant, so you can ask where a wallet's money went without writing a query. The full history is also sold as files on the Bitquery Data Store, as the Ethereum dataset.
This article is provided for informational and educational purposes only and reflects analysis of publicly available on-chain data as of the dates indicated. It does not constitute legal, financial, compliance, or investment advice, and nothing in it is a recommendation to buy, sell or hold any token or asset, or to use or avoid any exchange, wallet, bridge, protocol or service.
The findings describe transactions and balances observed on Ethereum on 18 November 2026. Attributions come from on-chain behaviour and Bitquery's own labels; they may be incomplete or incorrect and may be revised as more data becomes available. K3 Capital has not published a report, and its findings may differ from those described here.
References to K3 Capital, Liquity, ChainSecurity, Blockaid, ExVul, Morpho, Balancer, Curve, Pendle, Titan, BuilderNet, Ultrasound, Quasar, THORChain, Maya Protocol or any other named project, company or service describe what the public record shows and what those parties have said in public. They are not statements about any party's intent, security practices, solvency or conduct, and nothing here asserts that any named party acted unlawfully or negligently or knew of the events described. The wallets that took and moved the money are described by their behaviour and attributed to nobody.
Nothing herein should be relied upon as a definitive determination of fact. Readers should conduct their own independent verification before taking any action. The authors and publisher accept no liability for any loss or damage arising from reliance on this material. All trademarks and company names are the property of their respective owners.
Reported by Gaurav Agarwal for Bitquery Research, with AI tools; every figure was checked against the raw data.
FAQ
Was Liquity hacked?
Not as far as the transactions show. The drain hit sBOLD, a vault that K3 Capital runs on top of Liquity's Stability Pools. Liquity's troves opened, closed and paid out redemptions as normal. The weak point was how sBOLD priced its own shares.
How was the sBOLD vault drained?
The attacker borrowed BOLD through a Liquity trove and deposited it into sBOLD. It sent the vault a little WETH that the share price ignored. Then it called swap() to turn that WETH into BOLD that did count. It cashed out its shares at the higher price, and repeated the round several times in each transaction.
How much was stolen from K3 Capital's sBOLD?
The vault lost about $3.42M worth of BOLD, which was everything it held. It went from millions of BOLD to almost none in eight minutes.
What is a MEV bot?
A MEV bot is a program that watches pending Ethereum transactions and copies or jumps ahead of the ones that make money. It pays block builders to put its own transactions first. In the sBOLD drain, one MEV bot copied the attack and took 56% of the money.
Who got the money from the sBOLD hack?
A copycat MEV bot pulled out more than half and paid most of it to block builders, who received $1.78M in all. The wallet that first put the attack on chain and a wallet it funded took about $0.52M. Three other early wallets took most of the rest.
Did the attackers liquidate their own troves?
No. The first alerts said they did. During the drain, no trove was liquidated on any of Liquity's three markets. In each round, the attacker sent the WETH to the vault itself.
Is the sBOLD vault paused?
Yes. K3 Capital said on X that the contracts were paused, and the vault reports itself as paused. By then it held almost no BOLD.
Can the stolen funds be recovered?
Only in part, if at all. ETH cannot be frozen. Most of the attacker's share went to THORChain and to Maya Protocol, on its way to Bitcoin. Some ETH still sat in other early wallets, and in one fresh wallet, when we checked. An exchange can act if any of it reaches a deposit address.
Run the next investigation in minutes
Ask any question of the blockchain. Bitquery's AI Investigations agent traces funds, attributes wallets, and writes auditable reports across every major chain. Prefer to do it yourself? Connect the Bitquery MCP server to your AI assistant, pull the raw datasets, or query the chain directly with the Bitquery API.