A savings vault engineered to compound yield on Ethereum vanished in 480 seconds on November 18, 2026, leaving behind an empty smart contract and a stark portrait of automated exploitation. Between 12:25:23 UTC and 12:33:35 UTC, an attacker accompanied by copycat arbitrage bots drained approximately $3.42 million from the sBOLD savings vault—practically the entirety of its assets—leaving roughly 2,000 BOLD tokens in reserve.
The vault, designed and deployed by institutional liquidity manager K3 Capital as a yield-bearing wrapper around Liquity V2’s stability pools, collapsed under a barrage of 97 transactions spanning 37 Ethereum blocks. Rather than an isolated breach executed by a single hacker in a dark room, the event turned into a ruthless mempool feeding frenzy. A single copycat maximal extractable value (MEV) bot captured 56% of the drained funds ($1.91 million) and handed over $1.78 million of that loot directly to Ethereum block builders in priority gas fees to outpace competitors.
By the time blockchain security firm Blockaid broadcast an urgent alert to the public at 12:50 UTC warning of an ongoing drain, the vault was already an empty shell. When K3 Capital formally announced that it had paused the underlying contracts at 13:08 UTC, 35 minutes had elapsed since the final theft transaction was mined. The incident, now scrutinized across decentralized finance (DeFi) as the Ethereum sBOLD hack, exposes a critical fault line in how automated vaults calculate share prices when interacting with complex collateral mechanisms.
THE EIGHT-MINUTE COLLAPSE: AT A GLANCE
========================================================================
Total Funds Drained: $3,420,000+ (Virtually 100% of Vault TVL)
Vault Starting Balance: ~$3,423,000 BOLD
Vault Ending Balance: ~2,000 BOLD
Time Elapsed: 8 minutes, 12 seconds (12:25:23 – 12:33:35 UTC)
Total Exploit Transactions: 97 across 37 Ethereum blocks
Attacking Entities: 6 distinct wallets (1 originator + 5 copycats/bots)
Dominant Copycat Haul: $1.91 million (56% of total drained)
Bribes Paid to Block Builders: $1.78 million (52% of total drained)
Original Attacker Haul: ~$520,000 (Bridged to Bitcoin via THORChain/Maya)
Official Contract Pause: 13:08 UTC (35 minutes after vault was emptied)
========================================================================
The Architecture: How sBOLD Was Built to Harvest Liquity V2 Yield
Understanding why the vault broke requires unpacking what sBOLD was designed to do and how it interfaced with Liquity V2.
Liquity V2 operates as an overcollateralized lending protocol where borrowers deposit crypto assets—principally Wrapped Ether (WETH), Wrapped Staked Ether (wstETH), and Rocket Pool Ether (rETH)—to mint the protocol’s native dollar-pegged stablecoin, BOLD. Borrowing positions are isolated into individual debt positions known as troves. To protect the protocol from bad debt, Liquity V2 employs Stability Pools. Depositors place BOLD into these pools to act as an emergency liquidity cushion. When a borrower's trove becomes undercollateralized, the pool burns the necessary amount of BOLD to wipe out the debt and claims the borrower’s deposited collateral at a discount. Stability Pool depositors earn two streams of revenue: regular interest paid by active borrowers and liquidation gains acquired whenever an unhealthy trove is cleared.
Managing Stability Pool rewards manually is capital-inefficient for individual users. When collateral like WETH or wstETH is awarded during a liquidation, the user must repeatedly claim that collateral, sell it on a decentralized exchange, convert it back into BOLD, and redeposit it into the pool to compound their yield.
Intended sBOLD Workflow:
[User Deposits BOLD] ──> [sBOLD Vault (ERC-4626)]
│
┌───────────────────────┼───────────────────────┐
▼ (60%) ▼ (30%) ▼ (10%)
[wstETH Stability Pool] [WETH Stability Pool] [rETH Stability Pool]
│ │ │
▼ ▼ ▼
Liquidations: Liquidations: Liquidations:
Absorbs wstETH Absorbs WETH Absorbs rETH
│ │ │
└───────────────────────┼───────────────────────┘
│ Collateral rewards collected
▼
[Public swap() Function]
Swaps wstETH/WETH/rETH ──> BOLD via 1inch/DEX
│
▼
[Re-deposited to Pools to Compound Share Value]
To automate this pipeline, institutional asset manager K3 Capital launched sBOLD in mid-2025. Structured as an ERC-4626 tokenized vault—the standard interface for yield-bearing tokens on Ethereum—sBOLD pooled depositors' BOLD and routed the capital across Liquity V2’s Stability Pools in fixed allocations: 60% into the wstETH pool, 30% into the WETH pool, and 10% into the rETH pool.
In exchange for their stablecoins, depositors received sBOLD shares. The vault’s fundamental promise rested on a rising share price: as borrower interest accrued and liquidated collateral was swapped back into BOLD, the ratio of total underlying assets to total shares minted would increase. Users could redeem their sBOLD shares at any time through the standard redeem() function to receive more BOLD than they originally contributed.
Unlike Liquity V2’s immutable core protocol, K3 Capital’s sBOLD vault introduced several mutable and automated external dependencies. It relied on external oracle price feeds (specifically Pyth Network), an integration with DEX aggregators (such as 1inch v6) to process trades, and an open smart contract function named swap() that allowed keepers or external callers to convert harvested collateral tokens back into BOLD in exchange for caller incentives.
This swap() mechanism and the accounting rules surrounding assets waiting to be converted formed the structural foundation of the breach.
The Latent Warning: The May 2025 Audit Trail
The vulnerability that brought down sBOLD did not appear without warning. Eighteen months before the attack, formal security assessments had documented the precise friction point between vault share calculation and liquidation accounting.
On May 19, 2025, security firm ChainSecurity finalized a comprehensive security audit of K3 Capital’s sBOLD contracts. ChainSecurity's audit reviewed multiple iterations of the code, evaluating accounting correctness, reentrancy vectors, and interactions with Liquity V2 Stability Pools. In its findings, the auditors flagged a high-severity accounting vulnerability.
The auditor observed that an earlier version of the software could undervalue vault shares following liquidations. If the vault absorbed collateral from a liquidation event, that collateral sat in the vault's address before being converted back into BOLD. When calculating totalAssets()—the core variable in ERC-4626 vaults that dictates how many shares a new depositor receives for their money—the vault’s valuation algorithm omitted collateral balances below a predetermined threshold. In the audit’s theoretical models, this threshold was pegged at a value equivalent to roughly 7,500 BOLD.
THE THRESHOLD ACCOUNTING ARBITRAGE
──────────────────────────────────────────────────────────────────────────
Step 1: Collateral exists inside the vault below the threshold (e.g., < 7,500 BOLD).
Result: totalAssets() IGNORES this collateral.
Share price remains artificially depressed.
Step 2: Exploiter deposits large amounts of BOLD.
Mints sBOLD shares at the undervalued, unadjusted price.
Step 3: Exploiter calls swap().
The hidden collateral is sold on-market for BOLD.
The newly minted BOLD suddenly enters totalAssets().
Step 4: Share price jumps instantaneously.
Exploiter redeems sBOLD shares for full asset value.
Net profit is extracted directly from existing vault depositors.
──────────────────────────────────────────────────────────────────────────
In the May 2025 audit report, ChainSecurity noted that an actor could exploit this window by initiating a liquidation, waiting for the vault to receive collateral that remained temporarily uncounted or undervalued, minting sBOLD shares cheaply, calling swap() to realize the BOLD value, and then redeeming their shares at an inflated price. ChainSecurity marked this issue as resolved in "Version 3" of the codebase, stating in its executive summary that "the possibility of value extraction through self-liquidations has been mitigated". A secondary review conducted by Dedaub around the same period evaluated oracle reliance and BOLD liquidity without flagging unresolved critical exploits.
The conceptual tension, however, remained embedded in the contract logic: what happens if collateral arrives in the vault outside of a standard, predictable liquidation, and the share valuation function still fails to count it until swap() is executed?
Community chatter in the months leading up to the exploit hinted that independent researchers had encountered quirks in the sBOLD reward routing. On-chain messages and developer discussions following the incident revealed that at least one researcher had questioned Liquity’s community moderators about the vault’s accounting assumptions. Because sBOLD was an independent product maintained exclusively by K3 Capital and not part of Liquity’s core immutable architecture, the Liquity team directed inquiries elsewhere, while K3 Capital lacked an active, public bug bounty program with explicit disclosure frameworks.
Phase 1: The Deployment and Weaponization (12:14 – 12:25:22 UTC)
The opening move of the attack took place in broad daylight on the Ethereum blockchain, eleven minutes before the first dollar moved out of the vault.
At 12:14 UTC on November 18, 2026, a freshly funded address beginning with 0xdc0e broadcast a transaction deploying an unverified smart contract to the Ethereum mainnet. The address had been funded with gas money through cross-chain channels, leaving little trace on central exchanges. The deployed contract was tailored to interact simultaneously with three DeFi systems: Morpho Blue (a permissionless lending protocol), Liquity V2, and K3 Capital’s sBOLD vault.
Shortly after 12:20 UTC, the operator tested transaction paths. Rather than attempting a brute-force withdrawal or attacking the immutable core of Liquity V2, the attacker set up an automated multi-step loop within a single transaction frame.
CHRONOLOGICAL TIMELINE: THE GENESIS
==============================================================================
12:14:11 UTC | Address 0xdc0e deploys custom exploit execution contract.
12:18:45 UTC | Contract execution parameters calibrated across Morpho & Liquity.
12:24:50 UTC | Transaction constructed in private mempool / RPC relay.
12:25:23 UTC | Block 26,442,850: First drain transaction strikes the sBOLD vault.
==============================================================================
Initial reports from security firms hypothesized that the attacker initiated a massive flash loan to establish and deliberately self-liquidate a Liquity V2 trove. In that hypothesis, liquidating one's own loan dumped discounted collateral into the Stability Pool, depressing the vault’s share valuation.
Granular on-chain forensic deconstructions published by Bitquery later established a simpler, more surgical path. The attacker did not need to liquidate a trove at all.
The attacker used Morpho to take out a flash loan of 10,000 wstETH—a massive pool of capital borrowed and returned within the span of a single block. With this collateral, the attacker opened a fresh Liquity V2 trove and minted roughly 9 million BOLD. Now equipped with an enormous sum of liquid BOLD and a modest pool of WETH, 0xdc0e was prepared to execute an internal share-pricing squeeze on sBOLD.
Phase 2: The Strike at 12:25:23 UTC — The Precision Loop
At 12:25:23 UTC, transaction 0x8a9f... was mined on Ethereum mainnet. It initiated the first of many cycles that systematically extracted real capital from the vault.
Inside this single transaction, the attacker executed a sequence repeated eight consecutive times:
ANATOMY OF A SINGLE EXPLOIT LOOP
──────────────────────────────────────────────────────────────────────────
1. MINT / DEPOSIT:
Attacker deposits 9,000,000 BOLD into the sBOLD vault.
The vault issues sBOLD shares based on its current totalAssets().
2. DONATE COLLATERAL:
Attacker directly transfers 4 WETH into the sBOLD contract address.
Crucial Flaw: The vault's internal share pricing ignores this collateral
because it sits beneath the threshold or hasn't been converted.
3. TRIGGER SWAP:
Attacker invokes the public swap() function.
The vault sends the 4 WETH to an external DEX router, which trades it
for 7,159 BOLD.
The 7,159 BOLD is received by the vault and instantly recognized by
totalAssets().
4. INFLATION & REDEMPTION:
Because new BOLD entered the vault's balance sheet without new shares
being minted, the conversion rate of sBOLD shares increases instantly.
Attacker calls redeem(), cashing out their shares for the original
deposit PLUS an excess ~10,600 BOLD taken from other depositors.
5. REPEAT:
Loop repeats 8 times per transaction.
──────────────────────────────────────────────────────────────────────────
When the loop concluded, the transaction cleaned up its footprint: the attacker closed the Liquity V2 trove, repaid the 10,000 wstETH flash loan back to Morpho, and redeemed leftover BOLD through Liquity’s native collateral redemption mechanism for an assortment of WETH, wstETH, and rETH.
That single transaction netted the attacker approximately $23,000 in clean profit, extracted directly from the vault’s backing reserves. The Stability Pools themselves lost 24,318 BOLD in that initial burst.
Had the attacker been left alone to execute these loops methodically through private transactions, draining the full $3.42 million would have required several hours of disciplined execution.
Instead, the attacker made a critical operational mistake: they broadcast transactions through public nodes, alerting the automated surveillance bots that monitor Ethereum’s mempool.
Phase 3: The MEV Shark Tank and the Copycat Cascade (12:26 – 12:33 UTC)
Ethereum’s mempool functions as an adversarial arena. MEV bots run by searchers analyze every unconfirmed transaction, simulating execution against local forks of the blockchain to identify profitable state transitions.
When 0xdc0e executed its initial drains, automated bots detected the state change in the sBOLD vault. They did not need to understand the underlying economic design of Liquity V2 or K3 Capital’s vault; they merely decompiled the transaction input calldata, recognized that calling these contracts with specific parameters generated a net-positive balance of BOLD, replaced the recipient address with their own, and submitted the copied transaction with higher gas fees.
ESCALATION TIMELINE: THE 8-MINUTE RUN
==============================================================================
12:25:23 UTC | Block 26,442,850 | Attacker 0xdc0e executes first drain ($23,000).
12:26:47 UTC | Block 26,442,857 | Copycat MEV Bot enters the fray.
12:27:35 UTC | Block 26,442,861 | 4 additional bot-linked wallets begin firing.
12:29:10 UTC | Block 26,442,869 | Priority fee gas war drives bribes over $50k/block.
12:31:00 UTC | Block 26,442,878 | Vault TVL drops below $1,000,000.
12:33:35 UTC | Block 26,442,887 | Final drain transaction completes; ~2,000 BOLD left.
==============================================================================
At 12:26:47 UTC—barely 84 seconds after the original attacker’s opening transaction—the first copycat bot joined the fray. What followed was an aggressive, competitive sprint.
Over the next six minutes, six independent wallets engaged in an escalating gas war to drain the vault. The transactions spanned 37 consecutive Ethereum blocks, with bots repeatedly executing the deposit-donate-swap-redeem sequence at maximum velocity.
Because multiple bots were trying to call the exact same exploit function simultaneously, standard transaction fees were useless. If a bot's transaction arrived second in a block, the vault’s state would have changed, causing the transaction to revert and wasting computational fees. To ensure their transactions were placed at the very top of each block, the bots relied on MEV-Boost infrastructure, routing bundles directly to Ethereum block builders like Titan, BeaverBuild, and Builder0x69 accompanied by massive direct coinbase bribes.
The data reveals the scale of this MEV war:
- Total Transactions: 97 drain transactions succeeded across the eight-minute window.
- The Dominant Bot: A single copycat bot was responsible for 57 of the 97 successful transactions, capturing $1.91 million—representing 56% of everything stolen.
- The Block Builder Windfall: To win transaction placement, this dominant bot and its competitors surrendered almost all of their operational profits to the block builders. Out of the $3.42 million drained from the vault, $1.78 million (52%) was paid directly to Ethereum block builders as bribes.
- The Attacker's Share: The original exploiter (0xdc0e) and a closely linked secondary address (0x2324) were outpaced by the automated bots they had inadvertently awakened. Together, they secured approximately $520,000 before the vault was drained dry.
By 12:33:35 UTC, the vault's assets had collapsed from $3.42 million to roughly 2,000 BOLD. The smart contract remained online, but there was nothing left to take. The Ethereum sBOLD hack had concluded in eight minutes and twelve seconds.
WHERE THE $3.42 MILLION WENT
┌────────────────────────────────────────────────────────┬─────────────┐
│ Recipient Entity │ Dollar Haul │
├────────────────────────────────────────────────────────┼─────────────┤
│ Ethereum Block Builders (Coinbase Priority Bribes) │ $1,780,000 │
│ Dominant Copycat MEV Bot (Retained Payout Wallet) │ $240,000 │
│ Original Exploiter Group (0xdc0e / 0x2324) │ $520,000 │
│ Other Competing Copycat Wallets │ $880,000 │
├────────────────────────────────────────────────────────┼─────────────┤
│ Total Capital Siphoned from sBOLD Depositors │ $3,420,000 │
└────────────────────────────────────────────────────────┴─────────────┘
Phase 4: The Delayed Sirens and the 18-Minute Response (12:50 – 13:08 UTC)
In the decentralized ecosystem, off-chain awareness lags behind on-chain reality. While the sBOLD vault was stripped clean by 12:33 UTC, the broader DeFi market remained unaware for another quarter of an hour.
At 12:50 UTC—seventeen minutes after the last dollar had left the vault—blockchain security firm Blockaid published an alert on X (formerly Twitter):
"🚨 Blockaid detected an ongoing exploit on @k3_capital sBOLD on ethereum. Attackers flash-loaned wstETH, self-liquidated a Liquity V2 trove into the stability pools, minted cheap sBOLD shares, swapped the collateral, and redeemed other depositors' BOLD. ~$3.43M drained from the sBOLD vault so far across ~34 txs. Attack still ongoing."
Blockaid's monitoring bots had flagged the flurry of large-scale token burns, flash loans, and redemptions. However, because Blockaid’s heuristics clustered transactions around the initial waves, their alert estimated losses at $3.43 million across 34 transactions and characterized the exploit as "still ongoing". In truth, the exploit was already over; there were no remaining funds to drain.
The alert sparked panic across community channels. Liquity users rushed to Discord and Telegram to verify whether the core Liquity V2 protocol or the BOLD stablecoin itself had been compromised. Liquity contributors moved quickly to clarify the separation between the protocols: Liquity V2’s troves and Stability Pools were operating as intended without core code failures; the exploit was confined entirely to K3 Capital’s third-party ERC-4626 wrapper.
INCIDENT DISCLOSURE TIMELINE (UTC)
══════════════════════════════════════════════════════════════════════════════
12:25:23 | Exploit commences on Ethereum mainnet.
12:33:35 | Exploit concludes; vault drained of $3.42 million.
12:50:00 | Blockaid issues public alert on X; warns of "ongoing" exploit.
13:08:00 | K3 Capital issues official statement: contracts paused, probe opened.
══════════════════════════════════════════════════════════════════════════════
At 13:08 UTC—eighteen minutes after Blockaid’s public warning and thirty-five minutes after the drain ended—K3 Capital posted an official acknowledgment:
"We are investigating malicious activity involving sBOLD. Contracts have been paused."
K3 Capital invoked the emergency pause function embedded in the sBOLD vault contracts, freezing all deposits, mints, swaps, and redemptions. While the pause succeeded in preventing any secondary interactions, it acted on an empty vault. The team provided no loss estimates, offered no immediate timetable for reopening, and did not confirm whether any stolen assets had been successfully frozen or intercepted.
Phase 5: Divergent Forensics — Blockaid vs. Bitquery
In the hours following the freeze, security analysts began dissecting the incident, producing two distinct accounts of how the Ethereum sBOLD hack occurred.
THE FORENSIC DIVIDE
┌───────────────────────────┬──────────────────────────────────────────┐
│ Parameter │ Blockaid Initial Assessment │
├───────────────────────────┼──────────────────────────────────────────┤
│ Stated Mechanism │ wstETH Flash Loan + Trove Self- │
│ │ Liquidation into Stability Pools │
│ Transaction Count │ ~34 Transactions │
│ Status at 12:50 UTC │ "Ongoing" │
│ Primary Culprit Identified│ Wallet 0x2324 / 0xdc0e │
└───────────────────────────┴──────────────────────────────────────────┘
┌───────────────────────────┬──────────────────────────────────────────┐
│ Parameter │ Bitquery Post-Mortem Deconstruction │
├───────────────────────────┼──────────────────────────────────────────┤
│ Stated Mechanism │ Direct WETH Donation + Internal │
│ │ swap() Accounting Discrepancy │
│ Transaction Count │ 97 Transactions across 37 Blocks │
│ Status at 12:50 UTC │ Terminated 17 minutes prior │
│ Primary Culprit Identified│ 6 Wallets (Dominant MEV Bot took 56%) │
└───────────────────────────┴──────────────────────────────────────────┘
The divergence centers on whether Liquity V2 troves were actually liquidated.
Blockaid’s initial analysis suggested that the attacker used wstETH flash loans to open troves and then intentionally triggered their liquidation directly into the Stability Pools. Because Stability Pools absorb debt and distribute seized collateral to depositors (including the sBOLD vault), Blockaid reasoned that this influx of collateral temporarily distorted share pricing, enabling discounted minting. This explanation aligned with the theoretical attack vector outlined in ChainSecurity’s May 2025 audit report.
Bitquery’s transaction-level trace presented a different reality. By parsing the internal event logs and token transfers of all 97 transactions, Bitquery revealed that zero troves were liquidated during the drain.
The attacker used the wstETH flash loan and the Liquity trove solely to obtain the 9 million BOLD needed to buy into the vault at scale. Once inside, the attacker did not wait for a liquidation. Instead, they directly gifted 4 WETH to the vault contract. Because the vault’s totalAssets() accounting function suffered from an arithmetic blind spot that excluded un-swapped collateral balances beneath the threshold, the share price did not immediately reflect the added value. The moment the attacker triggered swap(), the WETH was converted into BOLD via 1inch, immediately registering on the vault’s balance sheet and inflating the share value.
The attacker did not break Liquity’s liquidation mechanics; they manipulated the vault's internal accounting by feeding it uncounted collateral directly.
Phase 6: Following the Money — Cross-Chain Laundering and Builder Windfalls
As researchers mapped out the exploit mechanism, on-chain investigators traced the flow of stolen capital.
The original exploiter, operating through address 0xdc0e and a secondary address it deployed (0x2324), extracted roughly $520,000 worth of crypto assets before being crowded out by the bots. The attacker acted quickly to obscure the trail:
- Conversion to Native Assets: The attacker redeemed the extracted BOLD stablecoins for raw Ethereum-based assets, including WETH, wstETH, and rETH.
- Cross-Chain Bridging: Rather than sending funds to centralized exchanges with Know Your Customer (KYC) requirements, the attacker routed the assets through decentralized cross-chain protocols. A significant portion was funneled into THORChain without transaction memos to convert the funds into native Bitcoin.
- Maya Protocol Routing: An additional 129 ETH was routed through Maya Protocol—a decentralized cross-chain liquidity network—and dispersed across five distinct Bitcoin addresses.
FLOW OF FUNDS: ATTACKER vs. BOTS
══════════════════════════════════════════════════════════════════════════════
[sBOLD Vault Reserves ($3.42M)]
│
├───────────────────────────────┐
▼ (15% - ~$520k) ▼ (85% - ~$2.90M)
[Original Attacker: 0xdc0e / 0x2324] [Copycat Bots & Searchers]
│ │
▼ ├────────────────────────┐
[Cross-Chain Swaps] ▼ (39% of Bot Haul) ▼ (61% of Bot Haul)
THORChain & Maya Protocol [Retained Bot Profits] [Block Builder Bribes]
│ (~$1.12M net) ($1.78M paid to
▼ Titan, BeaverBuild, etc.)
[5 Destination BTC Wallets]
══════════════════════════════════════════════════════════════════════════════
While the original attacker moved half a million dollars into Bitcoin, the remaining $2.9 million followed a different path.
The copycat bots were profit-maximizing programs operating on automated logic. In their rush to beat competing searchers, their algorithms escalated their direct coinbase transfers to block builders. As a result, $1.78 million of the depositors' stolen stablecoins ended up in the fee recipient wallets of Ethereum block builders.
These builders—neutral infrastructure providers in Ethereum’s proof-of-stake consensus that bundle transactions submitted via MEV-Boost—did not design the exploit or actively participate in the theft. Yet, due to the competitive structure of on-chain gas auctions, they became the largest financial beneficiaries of the hack, absorbing over 52% of all capital drained from sBOLD.
Phase 7: The Broader Contagion — A Tumultuous Day Across Web3
The drain of K3 Capital’s sBOLD vault was not an isolated crisis on November 18, 2026. The breach formed the centerpiece of a chaotic 24-hour cycle across decentralized finance that saw nearly $5 million vanish across three unrelated incidents.
DEFI SECURITY INCIDENTS: NOVEMBER 18, 2026
┌───────────────────┬──────────────┬─────────────┬───────────────────────────┐
│ Protocol Affected │ Network │ Losses │ Primary Root Cause │
├───────────────────┼──────────────┼─────────────┼───────────────────────────┤
│ K3 Capital sBOLD │ Ethereum │ $3,420,000 │ Vault share accounting / │
│ │ │ │ uncounted collateral lag │
│ Reental │ Polygon │ $454,000 │ Malicious/compromised │
│ │ │ │ admin proxy upgrade │
│ Gonka AI Bridge │ Cross-Chain/ │ $1,056,800 │ Compromised validator │
│ │ Ethereum │ │ BLS signing group key │
├───────────────────┼──────────────┼─────────────┼───────────────────────────┤
│ Total Day Losses │ Multi-Chain │ $4,930,800 │ Combined exploits │
└───────────────────┴──────────────┴─────────────┴───────────────────────────┘
On the Polygon network, tokenized real estate platform Reental suffered an exploit that drained approximately $454,000 from its stablecoin deposit contracts across two transactions. Blockaid’s monitoring revealed that an administrator address upgraded the proxy contracts for USDT and USDC tokens, altering the underlying logic to permit arbitrary withdrawals.
Hours later, Gonka AI’s cross-chain Ethereum bridge was compromised. An attacker acquired the platform’s Boneh-Lynn-Shacham (BLS) validator group key—the cryptographic secret used to verify and authorize cross-chain withdrawals. The attacker drained $884,800 in USDC and USDT from the bridge’s Ethereum reserves and minted $172,000 worth of unbacked Wrapped Gonka (WGNK) tokens, netting $1.06 million.
While these incidents stemmed from entirely different vulnerabilities—Reental suffered administrative proxy compromise, Gonka lost custody of its signing keys, and K3 Capital fell to smart contract accounting manipulation—their convergence within the same trading session triggered widespread risk aversion. Liquidity providers withdrew capital from newer lending markets, stablecoin yields fluctuated, and the security of ERC-4626 wrappers came under renewed scrutiny.
Technical Deconstruction: The Math Behind the Accounting Failure
To understand why the sBOLD contracts permitted this drain, one must examine the mathematics of ERC-4626 vault share minting and asset valuation.
In a standard ERC-4626 implementation, the relationship between shares and assets is defined by a basic ratio:
$$\text{Shares to Mint} = \frac{\text{Assets Deposited} \times \text{Total Share Supply}}{\text{Total Assets in Vault}}$$
$$\text{Assets to Redeem} = \frac{\text{Shares Burned} \times \text{Total Assets in Vault}}{\text{Total Share Supply}}$$
To prevent manipulation, the calculation of $\text{Total Assets in Vault}$ (totalAssets()) must accurately reflect all economic value held by the contract at the exact moment of execution. If a contract holds assets that are excluded from totalAssets(), the denominator in the share-minting equation is artificially small. Consequently, an incoming depositor receives far more shares than they are entitled to.
MATHEMATICAL SQUEEZE IN sBOLD
──────────────────────────────────────────────────────────────────────────
State A (Baseline):
Total Assets Recognized: 3,420,000 BOLD
Total Shares Minted: 3,420,000 sBOLD
Share Price: 1.0000 BOLD per share
State B (Collateral Donated but Uncounted):
Actual Assets in Vault: 3,420,000 BOLD + 4 WETH (~$12,000 value)
Recognized totalAssets: 3,420,000 BOLD (WETH ignored by threshold logic)
Share Price Calculated: 1.0000 BOLD per share
State C (Attacker Deposits 9,000,000 BOLD):
Deposited Assets: 9,000,000 BOLD
Shares Issued: 9,000,000 sBOLD (Minted at unadjusted 1.00 price)
Total Shares Now: 12,420,000 sBOLD
Recognized Assets Now: 12,420,000 BOLD
State D (swap() Converts 4 WETH into 7,159 BOLD):
New BOLD Added: 7,159 BOLD
New Recognized Assets: 12,427,159 BOLD
Total Shares: 12,420,000 sBOLD
New Share Price: 12,427,159 / 12,420,000 = 1.0005764 BOLD per share
State E (Attacker Redeems 9,000,000 sBOLD Shares):
Assets Returned: 9,000,000 × 1.0005764 = 9,005,187 BOLD
Attacker Net Profit: ~5,187 BOLD (Extracted from other depositors)
Remaining Depositors: Absorb the loss on their remaining shares.
──────────────────────────────────────────────────────────────────────────
In the sBOLD contract architecture, the function responsible for aggregating totalAssets() had to consult two different ledgers: the BOLD balance deposited in Liquity’s Stability Pools and the unliquidated/un-swapped collateral balances held directly in the vault.
Because swapping volatile collateral like WETH or wstETH back into BOLD incur gas costs and slippage, the contract was programmed to avoid micro-swaps. To save gas, it deferred small collateral balances, treating collateral below a set threshold as negligible noise until a sufficient balance accumulated to justify executing swap().
This threshold logic introduced an exploitable gap. By donating an amount of collateral that sat right near the threshold, an attacker could deposit millions in flash-minted BOLD at the unadjusted price, immediately trigger swap() to cross the threshold, force the vault to book the swapped BOLD into totalAssets(), and redeem their shares seconds later.
Because the attacker’s deposit represented the vast majority of all shares during those execution frames, the vast majority of the price jump accrued directly to them. When they redeemed, the BOLD returned to them exceeded their deposit, pulling capital straight out of the balances belonging to long-term depositors.
Repeating this operation across 97 transactions drained the entire $3.42 million reserve down to dust.
Systemic Vulnerabilities: The MEV Pipeline as an Accomplice
The escalation of the Ethereum sBOLD hack underscores a persistent challenge in Ethereum's transaction execution pipeline: the role of automated MEV searchers and block builders during smart contract breaches.
In traditional finance, an ongoing theft can be halted mid-stride by market circuit breakers, banking settlement delays, or administrative freeze orders. On a public, permissionless blockchain, an exploit transaction broadcast to the mempool becomes an open economic signal.
THE ADVERSARIAL MEMPOOL DRAIN PIPELINE
┌────────────────────────┐
│ Attacker broadcasts │
│ exploit tx to mempool │
└───────────┬────────────┘
│
▼
┌────────────────────────┐ Simulates tx on local fork;
│ MEV Searcher Bots │────> detects free balance extraction;
│ Scan Pending Pool │ replaces caller with bot address.
└───────────┬────────────┘
│
▼
┌────────────────────────┐ Bots outbid each other using
│ Priority Gas Auction │────> direct coinbase transfers
│ (PGA / MEV-Boost) │ (surrendering 90%+ of margin).
└───────────┬────────────┘
│
▼
┌────────────────────────┐ Block builder orders bundle
│ Ethereum Block Builder │────> at block apex to collect
│ (Titan, Beaver, etc.) │ bribe; validator signs block.
└───────────┬────────────┘
│
▼
┌────────────────────────┐
│ Vault drained faster │
│ than humanly possible │
└────────────────────────┘
When MEV bots detect an exploit, their algorithms are agnostic to whether the transaction is an arbitrage between two decentralized exchanges or the draining of a savings vault. The bot only evaluates profitability:
$$\text{Profit} = \text{Revenue Extracted} - \text{Coinbase Bribe Paid to Builder} - \text{Gas Execution Cost}$$
Because dozens of competing searchers run similar monitoring infrastructure, game theory dictates that the searchers will bid up their coinbase bribes to nearly 99% of the transaction’s extractable value to ensure their bundle is selected by the block builder.
During the sBOLD incident, this dynamic accelerated the theft beyond human response times. The original hacker did not drain $3.42 million; the hacker drained a few hundred thousand dollars, while competing algorithms stripped the rest of the vault in minutes to pay off block builders.
This dynamic raises difficult legal and architectural questions for Ethereum's consensus layer:
- Builder Neutrality vs. Complicity: Block builders argue they are neutral infrastructure pipelines that cannot and should not censor or validate the subjective intent of transactions. Yet, in this incident, builders collected $1.78 million of victims' capital in priority fees.
- The Failure of Emergency Pauses: K3 Capital’s multi-signature pause mechanism responded in 35 minutes. While 35 minutes is relatively swift for human coordination across a distributed team, it is meaningless in an environment where bots can empty a contract in eight minutes. On-chain security mechanisms that rely on off-chain human detection are fundamentally outmatched by mempool-scanning bots.
Core Protocol vs. Third-Party Wrappers: The Responsibility Divide
The Ethereum sBOLD hack highlights the ongoing operational friction between base-layer decentralized protocols and third-party yield aggregators.
Liquity V2 was engineered around immutability. Its core contracts cannot be modified, its governance is intentionally restricted, and its risk parameters are hardcoded into the protocol deployment. This design ensures predictability: users know the protocol's core rules cannot be changed by administrator keys or governance votes.
IMMUTABLE CORE vs. MUTABLE WRAPPER
┌───────────────────────────────────────┬──────────────────────────────────────┐
│ Liquity V2 Core Protocol │ K3 Capital sBOLD Vault │
├───────────────────────────────────────┼──────────────────────────────────────┤
│ • Immutable contracts (no upgrades) │ • Mutable parameters & pause keys │
│ • Zero external administrative roles │ • Managed by external multisig │
│ • Native Stability Pools │ • ERC-4626 Aggregation layer │
│ • Isolated trove debt accounting │ • External DEX swaps (1inch v6) │
│ • No third-party oracle for minting │ • Pyth oracle & custom math logic │
│ • STATUS: Uncompromised, 100% sound │ • STATUS: Drained, contracts paused │
└───────────────────────────────────────┴──────────────────────────────────────┘
However, raw protocols are often complex for ordinary capital allocators. The broader market frequently seeks "set-and-forget" products that compound yields automatically. Third-party firms step in to bridge this gap.
When K3 Capital built sBOLD, it introduced mutable components, external price feeds, automated DEX router integrations, and complex share-pricing algorithms on top of Liquity’s immutable foundation. While the marketing surrounding such vaults often emphasizes the security and resilience of the underlying protocol, the depositor’s true risk exposure is governed entirely by the weakest line of code in the wrapper.
Following the breach, retail depositors expressed frustration over this structural divide. Many users believed depositing into sBOLD was functionally equivalent to depositing into Liquity’s native Stability Pools, failing to realize that their capital was held by an external, mutable contract maintained by a third-party entity.
This incident demonstrates that an immutable, audited base protocol cannot protect users if the yield-bearing wrapper wrapping it contains accounting flaws.
Chronology of an Exploit: The Complete Timeline
To visualize the escalation of the breach, the complete lifecycle of the incident is mapped below:
THE COMPLETE sBOLD COLLAPSE TIMELINE
══════════════════════════════════════════════════════════════════════════════
MAY 19, 2025
• ChainSecurity publishes formal security audit of sBOLD codebase.
• High-severity finding notes vault share undervaluation risk tied to
uncounted collateral below threshold (~7,500 BOLD).
• Finding is categorized as "mitigated" in Version 3 of the code.
MID-2025
• K3 Capital deploys sBOLD on Ethereum mainnet.
• Deposits open across wstETH, WETH, and rETH Liquity V2 Stability Pools.
NOVEMBER 18, 2026 (MORNING UTC)
• 12:14:11 UTC | Address 0xdc0e deploys unverified exploit contract.
• 12:20:00 UTC | Attacker funds contracts and readies flash loans via Morpho.
THE DRAIN: 12:25:23 – 12:33:35 UTC (8 MINUTES, 12 SECONDS)
• 12:25:23 UTC | Block 26,442,850: First drain transaction strikes.
Attacker mints 9M BOLD, donates 4 WETH, calls swap(),
redeems inflated shares for $23,000 profit.
• 12:26:47 UTC | Block 26,442,857: Copycat MEV bot detects mempool pattern
and enters the execution queue.
• 12:27:30 UTC | Gas war escalates. 6 wallets compete across 37 blocks.
• 12:30:15 UTC | Dominant bot fires 57 successful drain calls.
Direct coinbase bribes to block builders surpass $1.7M.
• 12:33:35 UTC | Block 26,442,887: Final drain transaction executes.
Vault balance falls from $3.42M to ~2,000 BOLD.
THE AFTERMATH: 12:50 – 16:14 UTC
• 12:50:00 UTC | Blockaid issues public alert on X; reports ~$3.43M drained
across 34 transactions; calls exploit "ongoing."
• 13:08:00 UTC | K3 Capital pauses sBOLD contracts (35 minutes after drain).
• 13:30:00 UTC | Attacker 0xdc0e routes ~$520k through THORChain and Maya
Protocol to Bitcoin addresses.
• 16:14:00 UTC | On-chain state read at block 26,443,076 confirms vault
gutted; Bitquery launches block-by-block trace.
══════════════════════════════════════════════════════════════════════════════
What Comes Next: Unresolved Questions and the Path Ahead
As K3 Capital continues its internal investigation, several critical issues remain unresolved for depositors, auditors, and the broader DeFi ecosystem.
Restitution and Depositor Compensation
K3 Capital has not yet announced a formal compensation plan or reimbursement framework for affected depositors. Unlike decentralized protocols backed by vast venture treasuries, third-party vault curators often operate with lean balance sheets. Whether K3 Capital will absorb the $3.42 million loss through balance sheet equity, seek an insurance claim, or negotiate with the block builders and white-hat communities remains unconfirmed.
The Block Builder Ethical Dilemma
The fact that Ethereum block builders pocketed $1.78 million (over half of the stolen funds) has renewed calls for standard operating procedures regarding exploit-linked priority fees. While block builders have historically returned funds when explicitly requested during high-profile white-hat rescues, priority fees paid through competitive MEV-Boost auctions are usually distributed immediately to underlying validator pools as staking rewards. Unwinding those distributions across thousands of independent validators is technically and legally complex.
Code Audits and ERC-4626 Standards
The failure of sBOLD will likely spur revisions in how auditing firms assess ERC-4626 implementations that hold non-native or unliquidated assets. The assumption that an accounting vulnerability is "mitigated" because self-liquidation appears unprofitable overlooks direct asset manipulation, where an attacker bypasses protocol-level liquidations entirely and simply donates assets directly to trigger arithmetic thresholds. Security firms are re-evaluating audit guidelines for vaults that interact with external swap routers like 1inch to ensure that un-swapped balances cannot be exploited to artificially skew share prices.
On-Chain Circuit Breakers
The eight-minute duration of the drain proves that human-triggered multisig pauses cannot defend against automated exploits. If protocols are to survive in an adversarial MEV environment, future vault architectures must incorporate native, programmatic circuit breakers. These mechanisms could include:
- Hard limits on the percentage of total vault assets that can be withdrawn or redeemed within a single block or rolling time window.
- Dynamic withdrawal delays triggered whenever totalAssets() fluctuates beyond a predefined statistical threshold within a short block interval.
- Mandatory separation between share minting and share redemption, enforcing a minimum lockup of several blocks to neutralize single-transaction flash-loan manipulation.
The Ethereum sBOLD hack stands as a case study in modern DeFi vulnerability: an audited system running atop an immutable base protocol, dismantled in 97 transactions by automated bots, while the vast majority of the proceeds were claimed not by the attacker, but by the network’s own transaction ordering infrastructure. As investigations continue and the vault remains frozen, the incident offers a sharp reminder that in composable finance, a system is only as secure as the arithmetic that prices its shares.
Reference:
- https://bitquery.io/investigations/k3-capital-sbold-hack
- https://finance.biggo.com/news/0b4dcd5d-8ef7-41eb-af16-b8622d90cad3
- https://www.tokenpost.com/news/technology/60597
- https://finance.biggo.com/news/0b4dcd5d-8ef7-41eb-af16-b8622d90cad3
- https://www.liquity.org/blog/sbold---the-on-chain-defi-savings-account
- https://docs.liquity.org/v2-faq/bold-and-earn
- https://cryptobriefing.com/blockaid-k3-capital-sbold-exploit/
- https://thedefiant.io/news/hacks/blockaid-flags-3-9-million-in-k3-and-reental-exploits
- https://www.gfmreview.com/crypto/k3-capital-reental-lose-nearly-3-9m-in-separate-defi-exploits
- https://cryptobriefing.com/blockaid-k3-capital-sbold-exploit/
- https://www.coingecko.com/en/coins/sbold-by-k3-capital
- https://crypto.news/k3-capital-reental-defi-exploits-3-88-million/
- https://ourcryptotalk.com/news/k3-capital-sbold-exploit-3-43m-ethereum
- https://www.chainsecurity.com/security-audit/k3-sbold/
- https://ourcryptotalk.com/news/k3-capital-sbold-exploit-3-43m-ethereum
- https://bitquery.io/investigations/k3-capital-sbold-hack
- https://bitquery.io/investigations/k3-capital-sbold-hack
- https://tree.news/news/blockaid-detected-an-ongoing-exploit-on-k3-capital-sbold-on-ethereum-attackers-g4o90otp00sk
- https://pluang.com/en/news-feed/k3-capital-hentikan-kontrak-ethereum-sbold-setelah-peringatan-exploit-3-43juta
- https://crypto.news/gonka-bridge-sbold-vault-ethereum-exploits/
- https://finance.biggo.com/topics/Latest
- https://www.tokenpost.com/reporters/26?page=2
- https://www.kucoin.com/news/flash/k3-capital-s-sbold-vault-under-attack-3-43m-stolen
- https://curation.yearn.fi/report/yearn-yvusd/
- https://www.chainsecurity.com/docs/right-round-franco-sainas-thesis.pdf
- https://unchainedcrypto.com/k3-capital-pauses-sbold-contracts-blockaid-says-attackers-have-drained-3-43m/