Decentralized lending protocol Term Labs has become the latest DeFi platform to suffer an exploit, with an attacker withdrawing a total of $8.5 million in ETH and DAI by abusing the protocol's vault governance rules.

Term Labs is the creator of Term Finance, a decentralized lending protocol for fixed-rate ETH loans. The project set out to offer more predictable lending rates and highlighted a traditional quant team that included former Citibank and Morgan Stanley experts.

The team confirmed that the exploit affected some of its vaults, adding that the true impact is still being assessed.

We are aware of a governance exploit impacting Term vaults. We will share more details once it has been further investigated.

— Term Labs (@term_labs), August 23, 2026

Initial data show the attacker did not use a malicious exploit in the conventional sense. Instead, they relied solely on their governance influence, exactly as the protocol intended.

As of August 23, Term Labs held over $25 million in total value locked, according to DefiLlama. The protocol carried $3.92 million in active loans alongside a higher amount of collateral in its vaults. In total, Term Finance vaults held $12.25 million, DefiLlama data show, meaning the incident almost completely drained the protocol's lending capabilities.

The attack against Term Labs comes just days after other high-profile exploits, including Maya Protocol and a recent mint attack against The Sandbox.

How the attacker drained funds from Term Labs

On-chain data shows the attacker's wallets were funded with 2 ETH originating from Tornado Cash, an Ethereum mixing service that has appeared repeatedly in the funding trails of past crypto exploits. This type of initial funding has been linked to previous exploits carried out by DPRK hackers.

The attacker controlled four of the five drained vaults on Term Labs simply by holding 100% of the governance token.

Term Labs offered lending vaults similar to Morpho, where anyone could deposit funds and receive passive income. However, Term Labs also used Aragon as its governance platform, one of the longest-running open-source frameworks for building and running DAOs, and users could optionally wrap their vault deposits into governance tokens. To obtain governance tokens, users had to manually take the share tokens received from the vault and convert them into a special governance token.

By performing this second step, the attacker gained an unfair governance advantage and was able to use that influence to drain the vaults. Although the governance tokens held were worth only a few dollars, the attacker wielded outsized influence and moved all of the reserves from the vaults.

The attacker submitted a proposal on August 17 containing hidden actions that were not immediately visible to voters. Waiting periods of this kind function as timelocks, a common DAO safeguard intended to give participants time to review and veto pending changes before execution. After the six-day waiting period elapsed, the attacker changed vault parameters in a way that allowed funds to be drained from five USDC lending vaults.

After moving the funds, the exploiter parked them in a single known wallet, holding $1.6 million in DAI and around $6.9 million in ETH. The funds have not yet been mixed or moved, unlike other exploits in which mixing followed within the first hour. Because the address is publicly visible, the wallet can be monitored on-chain, and blockchain analytics firms and exchanges routinely flag addresses tied to stolen funds if movement or deposit attempts occur.

Why governance attacks are successful

Multiple Web3 protocols have built some type of governance mechanism, usually tied to holding specific tokens. The presence of whales, team allocations, or aggressive buyers means reserves, DAO treasuries, or other vaults can be attacked and drained.

The best-known precedent remains Beanstalk Farms, which lost around $182 million in April 2022 after an attacker used a flash loan, uncollateralized borrowing executed within a single transaction, to briefly acquire majority voting power and pass a self-serving proposal. Unlike Term's six-day waiting period, Beanstalk's emergency-proposal mechanism executed the attacker's changes immediately.

DAOs and Web3 organizations have varying rates of proposals, and not all proposals are understood by regular users. This means an interested party can propose an outcome in their own favor and successfully vote it through.