A vault was emptied in an eight-minute transaction race

K3 Capital's sBOLD savings vault lost about $3.42 million on November 18 as six wallets repeated an accounting attack across 97 Ethereum transactions, according to an onchain investigation by Bitquery. The analytics firm said the vault's BOLD balance fell from about 3.43 million tokens to almost nothing in roughly eight minutes. sBOLD is an ERC-4626 vault that accepts Liquity's dollar-oriented BOLD token and allocates deposits across Liquity V2 stability pools to earn yield. K3 Capital said it was investigating malicious activity and paused the contracts, but it has not published a full post-mortem, identified an attacker or stated whether depositors will be compensated.

The copycat bot captured the largest share

Bitquery found that one automated bot copied the attack and extracted about $1.91 million, equivalent to 56% of the drained value. The wallet that first placed the attack onchain, together with a second wallet it funded, ended with about $520,000, the firm said. CoinDesk independently reviewed the analysis and reported the same distribution. These are onchain attributions based on transaction timing, code similarities and fund flows; they do not establish who controlled the wallets or whether any participant knew the full context before submitting a transaction.

Related reporting: MEV bot intercepts $7.8M in rsETH after Safe module exploit

Block builders received about $1.78 million

The race for transaction priority transferred about 986 ETH, valued by Bitquery at $1.78 million, to the companies assembling the relevant Ethereum blocks. The copycat bot paid most of its proceeds for priority, with one inclusion payment reaching 44 ETH. Titan built most of the affected blocks, while BuilderNet, Ultrasound and Quasar built others, according to the analysis. Such priority payments are a normal part of maximal extractable value, or MEV, markets, where bots compete over transaction ordering. In this case, the opportunity came from copying a harmful transaction rather than an ordinary arbitrage. Those receipts should not be treated as confirmed profit: builders generally pass part of their block revenue to validators, and the available data does not establish what each retained. Bitquery also said it found no sign that the builders knew what the transactions were doing.

Analysts disagree on the precise exploit sequence

Bitquery's reconstruction says the attacker deposited BOLD, sent wrapped ether directly to the vault where it was initially excluded from share-price accounting, converted that ether into BOLD that was counted, and then redeemed shares at an inflated value. Early alerts described self-liquidations as part of the attack, but Bitquery said it found no Liquity trove liquidations during the drain. That discrepancy remains unresolved. Liquity has said its core protocol was not affected and described sBOLD as a separately operated vault.

An earlier audit does not settle the cause

ChainSecurity audited sBOLD in May 2025 and examined accounting correctness, stability-pool interactions and a potential value-extraction route involving self-liquidations. Its final report said the identified high-severity issues had been mitigated in the reviewed version and rated that version's security as high, while noting that audits cannot uncover every vulnerability. CoinDesk said it is unclear whether the November attack involved a related mechanism, whether the reviewed fixes were deployed or whether a separate accounting weakness was responsible. Until K3 Capital publishes its own technical account, the transaction analysis is strong evidence of the fund flows but not a final determination of root cause.

Sources

AI-generated editorial image; not a photograph of the reported event. Prepared with AI assistance and source verification.