Security
$951 Bought the Votes. The Vaults Did the Rest
Term Finance's Meta Vaults were not hacked. An attacker acquired voting control of a sparsely held governance token, passed proposals in the ordinary way, and the vaults transferred out $8.5 million — about 68% of everything depositors had put in — because that is what they had been instructed to do.
On Aug 23, Term Labs confirmed that a governance exploit had hit Term Vaults — the vault layer sitting on top of its fixed-rate lending protocol, Term Finance. By the time it was over, roughly 2,843 ETH (about $6.87 million) and 1.68 million USDC had left, the USDC promptly swapped for about the same amount in DAI. Call it $8.5 million.
That is around 68% of the $12.45 million depositors held in the vault product, and very nearly all of its approximately $8.8 million in ether. The security firms PeckShield and CertiK flagged the movement and traced it to a wallet beginning 0xD5183.
Now the number that makes this worth writing about eight days later. According to crypto.news, drawing on analysis from Defimon, the attacker spent roughly $951 acquiring the voting tokens needed to take control. Not $951,000. Nine hundred and fifty-one dollars — enough to command four USDC strategy vaults and about 91% of the Ethereum Meta Vault, then vote the contents out through the front door.
| Item | Reported figure |
|---|---|
| Spent acquiring voting control | ≈ $951 |
| Initial wallet funding | 2 ETH, sourced via Tornado Cash |
| Control obtained | 4 USDC strategy vaults; ≈91% of the ETH Meta Vault |
| Vault TVL before | ≈ $12.45M |
| Taken | 2,843 ETH + 1.68M USDC ≈ $8.5M (≈68%) |
What Term has said, and what it hasn't
Term Labs moved quickly on containment and was plain about it: "All Term Meta Vaults were shut down and DAO governance roles have been revoked. This shutdown is irreversible and permanently prevents further deposits. Withdrawals remain open." It attributed the incident to "a custom governance wrapper around the vaults," and said the underlying Term protocol and its direct borrowing and lending markets were unaffected, while noting it was still verifying scope. It has said it will "explore paths to address any remaining shortfall" — which is not a commitment to repay, and should not be read as one.
What Term has not said is how the attacker got the votes. As Cointelegraph noted, the company "has not confirmed how the attacker obtained voting control or which governance functions were used." The $951 figure and the majority-acquisition account come from third-party on-chain analysis, not from Term. That distinction matters and we are flagging it rather than burying it: the most striking detail in this story is currently unconfirmed by the only party in a position to confirm it. No technical postmortem has been published.
One more piece of the record. The vaults were built on Yearn V3 infrastructure, and Yearn moved to draw a boundary — the attack involved a custom governance wrapper, and the vector "does not apply to standard Yearn vault setups." On the facts available that appears to be accurate. It is also exactly what you would expect Yearn to say, and neither observation is a substitute for Term's postmortem.
Governance as an attack surface, priced
Strip out the vocabulary and this is a very old problem wearing new clothes. A vault was given an owner. The owner was a token. The token was cheap and thinly held. Therefore the vault was cheap.
Nothing was exploited in the sense that word normally carries. There was no reentrancy, no arithmetic overflow, no unchecked external call, no forged signature. Every transaction in the sequence was authorised by the governance system operating precisely as written. The attacker did not break the lock; the attacker bought the key from a shop that did not realise it was selling keys.
This is the same shape as the collateral-manipulation attacks that dominated August, and the resemblance is not coincidence. In both, an attacker identifies a variable the protocol trusts — a price, a vote — notices that the variable is cheap to move because the market setting it is shallow, moves it, and collects. The oracle version costs more to run because you have to buy real tokens on a real market. The governance version, in this case, cost $951.
The whole category, one incident
CertiK's August accounting, published today, puts governance losses for the month at $8.5 million. Term Finance is $8.5 million. The category and the incident are the same event.
Two readings are available and only one of them is comforting. The comforting one: governance capture is rare, a single event in a month with the highest incident count of 2026 so far. The other: a governance attack on a live protocol is now demonstrably a sub-$1,000 trade, it was executed successfully in August, and the sample is small because few people have bothered — not because it is difficult.
Which reading holds depends on how many DeFi products currently have a governance token with a thin float and real assets behind it. Nobody publishes that list. It would not be a hard list to compile, and there is every reason to think somebody already has.
Where it stands
The Meta Vaults are permanently closed. Withdrawals remain open for whatever is left, which for ether depositors is close to nothing. Governance roles are revoked, so the specific route in has been welded shut. The core lending protocol continues.
Outstanding, as of publication: no postmortem, no confirmed acquisition mechanism, no attribution, no recovery, and no repayment commitment beyond "explore paths." The funds' initial trace runs back through Tornado Cash, which usually means attribution will not arrive from on-chain analysis alone.
The Take
The lesson is not "audit your governance wrapper," though somebody should have. It is that a governance token with a small float and a large treasury behind it is a mispriced asset, and mispriced assets get arbitraged — that is the one law of this industry that never fails. Every protocol that has ever launched a low-float token and pointed it at real money has written an option, and the strike price is whatever it costs to buy the quorum. Term's was $951. The genuinely useful exercise for anyone running a DeFi product this week is not another code review; it is a single arithmetic question — what would it cost, at today's market depth, for a stranger to buy control of us — and then finding out whether anyone in the organisation is responsible for the answer. The most likely finding is that nobody is, which is how Term ended up here.