The Two Ether That Voted $8.5 Million Out of Term's Vaults

The Two Ether That Voted $8.5 Million Out of Term's Vaults

The smallest checkable fact in the Term Finance incident is the price of a yes. On Sunday an address pulled two Ether out of Tornado Cash, spending below five thousand dollars, and by the time the trades settled roughly $8.5 million had left Term's fixed-rate lending vaults for a single destination wallet beginning 0xD5183d8B. Set aside the usual reflex and read the receipts: no contract was broken, no oracle was corrupted, no audit surface failed. The money walked out the front door because the vaults' own governance voted it out. The code did exactly what it was told — which is the part that should worry depositors.

The receipts

On-chain, per PeckShield and CertiK, both of whom tracked the flow to the same address, the take breaks down as 2,843 ETH from the Ethereum Meta Vault, worth about $6.87 million at the moment of extraction, plus roughly $1.68 million of USDC from the USDC strategy vaults, swapped into about 1.6 million DAI before consolidation. The attacker never needed a bug to get there. The trace shows voting control of 100 percent over four of the five USDC strategy vaults and roughly 91 percent of the Ethereum Meta Vault — enough to pass their own proposals and instruct the contracts to move the money. Term Labs confirmed the incident in a single sentence: "We are aware of a governance exploit impacting Term vaults. We will share more details once it has been further investigated."

There the receipts stop and the gap begins. The controller of 0xD5183d8B is unattributed as of publication; the mixer cut the funding path precisely so the paper trail would end. Term had not, as of the reporting, published a technical postmortem naming the exact governance function the attacker used. The two-Ether seed is the documented entry point; whether more voting weight was assembled before the vote is an open question a postmortem still has to answer. The shape of the attack, though, is already clear.

Why control got cheap

Term's vaults — built on Yearn v3 infrastructure — govern withdrawals on-chain. In that architecture a vote is not won by convincing most of the community. It is won by outweighing everyone who bothered to show up. Participation across token-governed protocols typically runs in the single digits to low teens of voting weight; Forbes put it at 5 to 15 percent, with fewer than 1 percent of holders controlling roughly 90 percent of the power in many DAOs. When the quorum bar is low and turnout is thin, a concentrated position is not a minority that needs persuading. It is the majority. Control of eight and a half million dollars, bought with a seed under five thousand, is the arithmetic this whole class of attack now runs on.

None of this is new. It is newly cheap. The canonical version, Beanstalk in 2022, needed a flash loan — roughly $182 million of momentarily borrowed liquidity — to seize supermajority voting power inside one block. This summer's edition does not need the loan at all: Blockaid counted at least seven protocols across Ethereum, Solana, and Base between June 9 and August 6 losing about $22 million combined to governance takeovers. The catalogue runs from BonkDAO in July — about $20 million for the price of $4.4 million of BONK, roughly 1 percent of supply, executed with no delay between passage and payout — down to BarnBridge, where about $600 of locked BOND bought command of stale approvals that swept out $777,000. The Token of Power case needed only a fixed supply of 16,384 tokens and a one-transaction create-vote-execute path.

Term's weekend drain lands in that ledger. On DefiLlama's accounting, August had already logged 17 security incidents worth $18.8 million before this one; the $8.5 million pushes the month past $27 million, and it lifts the year's documented governance-attack total — five incidents, $25.1 million — past $33 million. The cost of seizing control keeps falling while the assets it controls do not.

The second incident, opposite ending

The detail that should change how a depositor reads this is not the mechanism. It is the difference between Term's own two incidents.

In April 2025, Term Finance lost roughly $1.6 million to the other failure class: a misconfigured price oracle on the tETH feed mispriced collateral and triggered abnormal liquidations. That was an internal error with a traceable path, which is why the recovery followed — 223.197 ETH was recaptured internally, another 333 ETH was secured through negotiation, and the team said it would cover the remaining hole from the protocol treasury. Depositors were, in effect, made whole.

This time the failure belongs to no one internal. An external operator funded through a mixer instructed the vaults, and the vaults obeyed because they were designed to obey a winning vote. The paper trail terminates at Tornado Cash, and the recovery math follows the identity of the loss: an internal decimal error can be refunded; a governance majority that won under the protocol's own rules can only be unwound if the thief chooses to hand the money back. Venus showed in 2025 that about $11.4 million of a $13.5 million loss can be recovered through governance action and negotiation, so the door is not shut. But the comparison that matters runs through Term's two incidents, not the industry's best case.

The break condition

In market terms this changes almost nothing. Ethereum trades near $2,447, up about 27 percent in five days and 32 percent in twenty (Ainvest market data), inside a crypto complex worth roughly $2.6 trillion. The $8.5 million is rounding. Against Term's reported $12.2 million in total value locked, most of it on Ethereum, the drain took the vault book in one motion — the protocol's core fixed-rate lending positions were not the affected surface.

The judgment that changes is narrower and sharper. "Audited" described the code, and the code held. It said nothing about the constitution: the vote mechanics, the quorum threshold, the delay between a passed proposal and its execution, whether voting weight is fixed at a past snapshot or counted on voting day itself. Every mitigation in this attack class is a boring parameter — a timelock of 24 to 48 hours between approval and execution, a quorum too high for a weekend's turnout to outbid, voting power locked to a block before the proposal was created. One of them was evidently missing at Term, and the postmortem will name which.

That sentence is the break condition for this read. If the missing guardrail was a timer that simply never got set, the cause is a config of a transitional governance window — the migration of TERM governance control to a Governor contract was scheduled, and interim arrangements carry interim risks — and the fix is mechanical. If the vaults were governed exactly as designed, step for step, then the design was the vulnerability, and these receipts read differently for every worthwhile vault protocol behind them. The wallet, the amounts, the percentages are public and checkable in minutes. The parameter is not — yet.