Term Labs Vaults Compromised Through Governance Exploit, $8.5 Million Lost
A controlling interest in the vaults was acquired for less than one ETH, and even that was excessive given the circumstances.
On August 23, Term Labs experienced a major loss of about $8.5 million from its vaults, following a governance attack that did not require any vulnerability in the core vault code.
The attacker secured 90.7% of the voting supply in the ETH Meta Vault with minimal capital, and gained complete voting control over five separate USDC vaults.
With a single deposit, the attacker’s wallet obtained majority control over all five USDC vaults and nearly all voting power in the ETH Meta Vault, a situation enabled by the lack of participation from other tokenholders.
Six days after these deposits, the attacker’s proposal was ready for execution as scheduled.
The initial actions taken set both the cooldown and expiration of the Zodiac Delay module to zero, eliminating the intended delay on high-risk transactions.
The process was systematic: Assets were withdrawn from legitimate strategies, an attacker-controlled strategy contract was introduced, assigned an unrestricted debt ceiling, and the recalled funds were sent to it.
The vaults lost around 2,843 WETH and 1.68 million USDC as a result.
Term Labs acknowledged that a governance vulnerability was at fault.
AMLBot noted in an August 23 update that the attacker’s funds remained unmoved at that time.
When governance is left unattended and timelocks are the sole security, do those safeguards actually provide any real protection?
Credit: The Block, Pyro, CD Security, PeckShield, Term Labs, AMLBot, Defimon, CertiK, Yearn, crypto.news
Defimon first identified the incident on August 23, with a Decurity bot publishing relevant transaction hashes and attacker addresses.
Defimon detailed the apparent technique before any official comment from Term Labs: The attacker "cheaply acquired a majority of a sparsely-held DAO governance token, then passed malicious proposals to seize control of Term's vaults."
Term Labs issued a preliminary response without providing loss figures or technical context: "We are aware of a governance exploit impacting Term vaults. We will share more details once it has been further investigated."
CertiK tracked the address holding the proceeds and assessed the total loss at about $8.5 million.
PeckShield’s follow-up reported a drain of about 2,843 ETH and 1.68 million USDC, with the USDC converted into DAI. PeckShield also highlighted that the attacker's initial funding came from 2 ETH via Tornado Cash.
Yearn later clarified that while Term’s vault contracts used Yearn V3 infrastructure, the vulnerability was due to a custom governance wrapper and did not impact standard Yearn vaults. Yearn confirmed that their regular vaults remained unaffected.
In a subsequent update, Term Labs announced that all Term Meta Vaults were permanently closed and their DAO governance rights revoked, preventing new deposits but allowing withdrawals to continue.
By then, independent monitors had already charted the key transaction routes, identified the attackers, tracked consolidation addresses, and publicly calculated the damages.
Term Labs did not address how a wallet with minimal financial commitment could acquire such overwhelming governance control.
If public researchers could reconstruct the on-chain sequence in a few hours, why did Term’s official statement by that evening offer less detail than was already available?
The Core Flaw
The exploit was not rooted in a defect in the underlying vault contracts.
Rather, it was the result of governance math and an authorization model broad enough to allow a large-scale withdrawal via governance.
Term’s Strategy Vaults were governed using Aragon TokenVoting, with voting power separated from regular vault shares.
The governance token was a wrapper for vault shares. Users needed to both deposit into a strategy vault and wrap their shares to receive voting rights. Few opted in, leaving most deposits without voting power. As a result, almost no one participated in governance.
On the ETH Meta Vault, the total governance token supply stood at just 0.5352. The attacker held 0.4852, about 90.7%, after depositing roughly 0.5 ETH and converting the shares.
USDC vaults showed a similarly small governance token supply; reports indicate the attacker held all active voting power in four out of five affected USDC vaults.
The protocol settings were not especially risky on their own: A 50% support threshold, a 5% minimum participation, and a voting period of just over six days.
Yet when almost all voting power is consolidated in one wallet, thresholds lose meaning.
Additionally, the minimum proposer threshold was set to zero, so anyone could submit a proposal without holding any voting power.
The attacker's proposal was titled “Veto strategy vault parameter change,” using the same format as routine updates: “Vote YES to VETO the curator’s proposed vault parameter changes. Otherwise, the transaction will become executable when this proposal expires.”
To casual observers, it appeared to be a standard veto item.
However, the proposal actually bundled 17 separate actions.
The first three actions reconfigured the Zodiac Delay module: They set the cooldown and expiration periods to zero, and enabled an attacker-controlled executor.
The remaining actions withdrew funds from four legitimate ETH strategies, added the attacker's own strategy (“Fixed Recipient WETH Exit Strategy”), set its debt ceiling to the maximum possible value, and transferred the vault’s entire balance.
The key issue is this: Low turnout made it easy to buy governance, but the governance system itself held enough authority to override its own safeguards.
The very process meant to oversee could also remove the restraints on itself.
It remains unclear whether this authority was intentional, a configuration oversight, or a separate authorization lapse.
No evidence has emerged of reentrancy, oracle manipulation, or private-key theft.
The attacker simply recognized that governance was neglected and that the controls could be circumvented by governance action itself.
If a delay module can be turned off by the same body it's meant to slow, what is it truly protecting against?
The Attack Sequence
The attacker withdrew all funds in just two transactions, spaced about 22 minutes apart, following the same governance-based method.
At around 06:25 UTC on August 23, the ETH Meta Vault proposal became executable. Moments later, the attacker called executeProposal().
The transaction pulled capital from four strategies, registered the attacker's strategy, gave it an unlimited debt ceiling, and drained the WETH from the vault.
About 2,841.74 WETH was extracted from the ETH Meta Vault.
Proposal Execution Exploit Transaction 1: 0xd354a15b15cb73d30908f411aee3f795ec86737a4d080e9a818ac4d6d3014129
At 06:47 UTC, a second attacker wallet followed the same approach with five USDC vaults in one transaction, draining around 1,679,639 USDC.
Proposal Execution Exploit Transaction 2: 0x9f273f9a5a20c2fc957b06bbfa45db486390eede4a7f44fbe1a2eb6744c2e8a0
Attacker Wallets: 0xa908b3472d76e7744bab0a5911768a4a6300612b 0x686457a7468b9b31c5dba43b1b16077b48520691
AMLBot observed that both wallets were funded with roughly 1 ETH each from Tornado Cash before the attack.
While this makes attribution more challenging, it does not directly reveal the attacker’s identity or organizational ties.
The stolen USDC was exchanged for DAI. The ETH/WETH and DAI proceeds were then consolidated at a single address:
Consolidation Address: 0xD5183d8BfC65a50863C62aF2538198A8288FFc13
According to monitoring reports, this address held about 2,843 ETH and 1.68 million DAI, totaling approximately $8.5 million in losses.
Following the exploit, 300 ETH was moved from the Consolidation Address to another wallet and subsequently withdrawn via Tornado Cash.
Wallet used for cash out via Tornado Cash:
0xC14007663A5bb9F13d4d2AEE8c6FE9075eF1d83e
Tornado Cash Transactions:
Three transactions are visible here
At present, the majority of the stolen funds remain together and traceable on the blockchain.
Whether the initial transfer was a laundering attempt, a test, or something else is still unclear.
Most funds are still visible on-chain, but visibility does not guarantee recoverability. What protections did the system really offer?
Governance Participation: A Critical Weakness
Term's Strategy Vaults were ERC-4626 tokenized vaults constructed on Yearn V3 architecture.
Yearn explained that the breach exploited Term’s custom governance wrapper and had no impact on Yearn’s standard vaults.
This distinction, however, may provide little comfort to depositors.
The contracts in question executed actions exactly as authorized by the on-chain proposal: they recalled WETH from four strategies, introduced the attacker’s strategy, assigned an unlimited debt ceiling, and transferred all funds.
The root problem was not an unauthorized vault call, but a definition of authorization that allowed this scenario.
While audits can detect technical bugs and dangerous permissions, they cannot compel passive holders to participate in governance, nor can technical soundness offset a governance token with such low active supply that control can be bought for a pittance.
Here, the risk was not in vault design, but in the ease and cost of acquiring effective control—and in the fact that a single proposal could change the rules that were supposed to limit it.
Term Labs encountered a significant issue before. In April 2025, a decimal error in the tETH oracle caused mispricing and resulted in about 918 ETH worth of unintended liquidations.
Term Labs attributed that event to an operational mistake, not a smart contract bug. The protocol managed to recover about 556 ETH, reducing the final loss to 362 ETH ($650,000), with assurances of full reimbursement to affected users.
Sixteen months later, the point of failure shifted from the oracle to the governance layer itself.
Following the recent incident, Term Labs closed all Meta Vaults, removed their governance capabilities, and made them permanently inaccessible to new deposits, while withdrawals stayed open.
The team maintained that, thus far, the core Term protocol and its direct lending/borrowing products were not impacted, though the investigation continued.
At the time of reporting, Term Labs had not released a comprehensive technical explainer, full postmortem, or any restitution plan for the $8.5 million loss. The attack was carried out using the governance authority the system permitted.
This does not mean such an outcome was inevitable, nor that mitigation is impossible.
It does, however, raise a critical question: If governance can disable the very checks it is supposed to operate under, what did the audit verify, and what remains for users to rely on?
A small amount of capital determined the fate of millions in depositor funds.
Term Finance’s loss of approximately $8.5 million was not due to a catastrophic code bug, but to near-total lack of governance engagement.
In all five USDC vaults, the attacker held the entire voting token supply; in the ETH Meta Vault, they controlled 90.7% of votes.
The malicious proposal was indistinguishable from routine updates in appearance, and its first steps removed the very delay meant to slow such actions.
The attacker needed only six days, a proposal disguised as standard business, and a delay module that could be reconfigured by governance itself.
Term Labs has since closed the Meta Vaults, removed their governance rights, and restricted new deposits while keeping withdrawals available.
As of the last available updates, Term Labs had not clarified why governance was permitted to change its own restrictions.
The $8.5 million loss is notable not for the sophistication of the attack, but because it highlights a recurring pattern in DAOs: the cost of low engagement.
Decentralization is only meaningful if there is active participation—otherwise, security is an illusion.
If control of a DAO can be bought for a few dollars’ worth of tokens, what does it mean to be “decentralized,” and who is really being protected?
Get new scam files the moment we publish them — usually 2–3 emails a week.