Term Finance: The Governance Attack That Beat a Timelock

At 06:25 UTC on Sunday, 23 August 2026, a single Ethereum block (number 25,816,049) settled a transaction that quietly emptied most of a lending protocol’s vaults. There was no reentrancy bug, no oracle glitch, no leaked private key. The attacker followed the rules. They had accumulated enough of Term Finance’s vault governance token to vote themselves control of the funds, and then they did exactly that. By the time Decurity’s monitoring bot noticed the outflows, roughly $8.5 million had left the building.

What sets Term Finance apart from the average DeFi hack is not the size of the loss. It is that Term had done almost everything the textbook tells a protocol to do. The vaults ran on Yearn V3, some of the most scrutinized infrastructure in decentralized finance. Governance actions carried a seven-day timelock. Liquidity providers held veto rights. The system even separated the “manager” role from the “governor” role so that no single key controlled every lever. And it still lost about 68% of its vault deposits to an attacker who bootstrapped the entire operation with two ETH, then worth under $5,000.

This is a forensic post-mortem of that attack: how it worked, why each guardrail failed, where the money went, and why Term was only the latest and best-defended casualty of a summer in which at least eight DAOs were voted out of their own treasuries. Governance attacks are no longer a curiosity. They are an exploit class with a repeatable playbook, and Term Finance is the clearest proof yet of the uncomfortable lesson underneath it: in 2026, the most dangerous vulnerability in a protocol is often the governance layer wrapped around perfectly good code.

What Term Finance Was, and What Its Vaults Held

Term Finance is an Ethereum-based lending protocol built around fixed-rate, fixed-term loans, an attempt to bring the predictability of a bond market to on-chain credit. Alongside its core lending markets, it ran a set of yield products called Term Meta Vaults, or Strategy Vaults, that pooled user deposits and routed them into lending and staking strategies. Those vaults were built on Yearn V3, the widely used vault framework, with a custom governance layer that let token holders steer each vault’s parameters. Depositors received vault tokens such as tmvETH, a claim on a pool of ETH being put to work chasing yield, the sort of staking-and-lending return we unpacked in our guide to validator economics in 2026.

The numbers were modest by blue-chip standards. According to The Block, Term held about $25.8 million in total value locked across the protocol before the attack, with roughly $12.45 million of that sitting in the vault products and about $8.8 million of the vault money on Ethereum. That scale matters to the story. A small, specialized protocol tends to have a small, specialized governance token, and a small governance token is a cheap thing to corner.

Term was not a fly-by-night project, either. In April 2025, an oracle error triggered about 918 ETH in unintended liquidations; the team recovered 556 ETH, cut final losses to 362 ETH, and reimbursed affected users, per Cointelegraph. That history of making users whole is worth holding onto, because it frames the central question the August attack left behind: when the loss comes not from a bug but from the protocol’s own voting system working as designed, who is on the hook?

The Attack, Minute by Minute

The exploit did not happen in a single dramatic transaction. It was a slow accumulation followed by a fast execution, the pattern a16z crypto researchers call surreptitious accumulation: buy quietly, behave like any other holder, and strike only once you hold enough votes. Here is the sequence pieced together from on-chain data and the confirmations that followed.

The detection was fast and the response decisive, but both came after the money was already gone. Decurity’s Defimon bot flagged the outflows publicly within minutes, and The Crypto Times pinned the execution to block 25,816,049 at 06:25 UTC. PeckShield and CertiK confirmed the loss shortly after. Term Labs then said it had “irreversibly shut down all Term Meta Vaults and revoked their DAO governance roles, preventing further deposits while keeping withdrawals open,” per Cointelegraph. The core lending and borrowing markets, the company said, were unaffected.

A Few Thousand Dollars for an $8.5 Million Door

The economics are the whole story. The attacker seeded the operation with two ETH, worth under $5,000 at the time, pulled from Tornado Cash to obscure its origin. Because Term’s vault governance token was so thinly held, that was enough to buy a controlling stake. A post-mortem summary published by exchange KuCoin put one decisive purchase at roughly $951, enough on its own to secure about 90.66% of the votes in a single pool. From there the attacker climbed to complete control of four of the five USDC vaults and about 91% of the Ethereum Meta Vault, according to The Block.

Crypto Briefing put the mismatch plainly: “The Term Finance exploit is a textbook example of what happens when voting power is too cheap or too concentrated. An attacker spending 2 ETH to control vaults holding millions of dollars in assets suggests that the cost of acquiring governance influence was wildly misaligned with the value of the assets those votes could control,” the outlet wrote.

That misalignment has a formal shape. In their canonical 2022 paper, a16z crypto researchers Pranav Garimidi, Scott Duke Kominers and Tim Roughgarden model a governance attack as a simple inequality: an attacker’s profit equals the value of the attack minus the cost of acquiring voting power minus the cost of executing it, and for a protocol to be safe that number has to stay negative. Term Finance inverted it. The value on the table was around $8.5 million; the cost of the votes was two ETH; the cost of execution was a handful of transactions. Profit was not just positive, it was enormous. By that measure Term was insecure by construction, and had been long before anyone attacked it.

The Timelock That Was Not a Lock

Here is the part that should unsettle anyone who has ever pointed to a timelock as proof a protocol is safe. Term had one. Vault governance transactions were subject to a seven-day delay, and liquidity providers could vote to veto a queued transaction. Term even separated the manager role, which ran auctions, from the governor role, which set risk parameters and protocol configuration, exactly the kind of role segregation security auditors ask for. On paper, Term’s governance looked responsible.

So why did none of it matter? Because a timelock is not a lock. It is a waiting period, and a waiting period only protects you if someone with both the power and the incentive to act is watching during the delay. Once the attacker held close to 100% of the voting power, they were effectively the only party with standing to veto their own proposal. The seven-day window became a formality: the one entity that could block the theft was the entity committing it. Every other holder was either absent, unaware, or outvoted.

Vitalik Buterin warned about precisely this failure mode years ago. Timelocks, he wrote in his 2021 essay on coin voting, “are more like a paywall on a newspaper website than they are like a lock and key,” a nuisance that slows a determined actor down without stopping them, unless there is an active, empowered defender on the other side. Term’s veto right assumed such a defender existed. Governance in practice is a spectator sport, and most token holders never see a proposal, let alone move to block one, the same passivity that makes malicious signing prompts so effective against users of even careful wallets like those in our MetaMask, Phantom and Rabby comparison. A guardrail no one is standing behind is not a guardrail. It is decoration.

The Code Was Fine; the Governance Was the Exploit

It is tempting to blame the infrastructure, but the infrastructure held. The vaults ran on Yearn V3, and Yearn was quick to clarify that the standard vault code was never the problem. The exploit, it said, “involved a custom governance wrapper and the attack vector does not apply to standard Yearn vault setups.” In other words, Term had wrapped its own governance logic around Yearn’s audited vaults, and that wrapper, not Yearn, was the soft spot.

Both firms that investigated agreed on the nature of the flaw. As Crypto Briefing summarized, PeckShield and CertiK “both confirmed the exploit, noting it targeted the voting mechanics rather than any flaw in the underlying smart contract code.” Nothing was overflowed, reentered, or mispriced. The contracts did exactly what they were programmed to do: they counted the votes, saw a passing proposal, waited out the timelock, and executed. The bug, if you can call it that, was that the rules themselves let a $5,000 stake command an $8.5 million payout.

This is what makes governance attacks so hard to defend against with the usual tools. A code audit checks whether the software does what the developers intended. It cannot tell you whether what the developers intended is safe against a well-capitalized voter, because from the contract’s perspective the attacker is not an intruder; they are a legitimate participant exercising a right the protocol granted them. a16z’s researchers call this the indistinguishability problem: a market cannot tell an attacker accumulating tokens from an enthusiast doing the same thing, since “both are willing to buy large quantities of tokens at increasingly high prices.” The exploit hides in plain sight, priced into the order book.

Two ETH In, $8.5 Million Out: Following the Money

The money trail is short and professional. The seed capital, two ETH, came out of Tornado Cash, the sanctioned mixer that remains the default on-ramp for attackers who want to sever the link between their funding source and their on-chain identity. On the way out, the attacker took 2,843 ETH, worth about $6.9 million with ETH trading near $2,400 at the time per CoinGecko, plus 1.68 million USDC.

That USDC did not stay USDC for long. The attacker promptly swapped it for roughly 1.68 million DAI, a move with a clear logic: USDC is issued by a centralized company that can freeze specific addresses on request, while DAI is far harder to claw back. It is the same instinct that sends bridge-hack proceeds through a familiar laundering sequence, the trail we followed in where stolen crypto goes in 2026: convert anything freezable into something that is not, then let it cool.

Ironically, the transparency of the chain is why we know all of this so precisely. Decurity’s Defimon bot caught the outflow in near real time, and blockchain forensics firms mapped the flows within hours. On-chain visibility is a genuine strength of DeFi; the tragedy of Term is that visibility after the fact does nothing to reverse an execution that was, by the protocol’s own rules, entirely valid. You can watch the vault drain in real time and be powerless to stop it, because the transaction draining it is not a hack. It is a vote.

Term Was Not Alone: The Summer 2026 Takeover Wave

Term Finance did not happen in isolation. Security firm Blockaid counted at least seven governance takeovers between 9 June and 6 August 2026, draining a combined $22 million across Ethereum, Solana and Base, and that tally closed before Term, on 23 August, became the eighth and one of the largest. The playbook barely changed from case to case: find a protocol with a thinly held governance token and weak execution controls, buy a majority for a pittance, and vote the treasury into your own wallet.

Blockaid’s list did not stop there; it also named Panther, taken over through its upgrade path, and Unicly, where the attacker flash-borrowed the voting power outright, per the firm’s breakdown. Seen together, the cases show a maturing craft. Token of Power in June was crude, minting ten billion new tokens to itself in one unguarded transaction. BonkDAO in July was bigger but still relied on a near-empty voting pool, draining about $20 million after the attacker spent roughly $4.4 million to clear quorum. By the time it reached Term in August, the technique had been refined against a target that had actually tried to defend itself.

Why Coin Voting Keeps Losing This Fight

Strip away the specifics and every one of these attacks rests on the same foundation: one-token-one-vote governance. A governance token, as Buterin has argued, bundles two very different things, an economic interest in the protocol and the right to steer it, and those two rights are trivially easy to unbundle. Someone can buy the votes without caring about the protocol’s long-term value, precisely because they intend to destroy that value and walk away with the treasury. The token market cannot price that intent.

Concentration makes it worse. A 2022 Chainalysis study of major DAOs found that across ten of them, less than 1% of holders controlled roughly 90% of the voting power. Turnout, meanwhile, is chronically low; many proposals pass on the participation of a tiny sliver of holders. That combination, concentrated supply plus voter apathy, is not so much a bug the attackers exploit as the water they swim in. When almost nobody votes, almost nobody needs to be outvoted.

Even the people who championed on-chain governance now speak about it with a certain fatigue. “We spent the last 10 years rediscovering the hard way that direct democracy is a bad idea,” a16z crypto general partner Ali Yahya told Fortune in June 2026. Term Finance is that sentence rendered as a crime scene: a small, disengaged electorate, a token that traded for pocket change, and a single motivated actor who understood that in a low-turnout system, buying the quorum is cheaper than earning it.

What Would Actually Have Stopped It

None of this is unfixable, and the fixes are well understood. Blockaid, a16z and OpenZeppelin have all published overlapping checklists, and most of the items would have blunted or stopped the Term attack outright. The common thread is to make voting power expensive to acquire, slow to weaponize, and impossible to convert into fund control through a single proposal.

• Snapshot voting power from a past block. If a proposal counts only the tokens held before it was announced, an attacker cannot buy the votes after the fact to pass their own measure.
• Raise quorum and proposal-deposit thresholds. Requiring a meaningful share of supply, and a real deposit, to propose or pass anything prices out the pocket-change takeover.
• Put a real timelock behind an empowered veto. A delay is only useful if an independent party, a security council or guardian multisig rather than just token holders, can actually cancel a malicious transaction during the window.
• Ring-fence the dangerous powers. Minting, treasury transfers and contract upgrades should sit behind extra controls, so no single governance vote can reach them.
• Reduce the free float. Staking, vesting and lockups shrink the tradable supply an attacker can corner, and give the token standalone value beyond pure voting.
• Monitor concentration in real time. Term’s outflow was flagged within minutes; the lesson is not that monitoring failed, but that alerting without an automated circuit-breaker arrives too late.

Note what is missing from that list: a better code audit. Every one of these measures is a governance-design choice, not a bug fix. That is the reframing Term forces on the industry. Security is not only a property of the code that moves the money; it is a property of the system that decides when the money moves.

Is a Governance Attack Even a Crime?

Here the story turns genuinely strange. When an attacker breaks into a system, the law has an easy time calling it theft. When an attacker uses the system exactly as built, the picture blurs. The defining precedent is Mango Markets. In 2022, Avraham Eisenberg drained about $110 million from that protocol through what he called a “highly profitable trading strategy,” was convicted of fraud and market manipulation in 2024, and then, in May 2025, watched a US judge vacate those convictions. Judge Arun Subramanian’s reasoning cut to the heart of the matter: Mango’s contracts were permissionless and executed automatically, which made it hard to say Eisenberg had lied to anyone or done anything the code forbade.

Apply that logic to Term. The attacker made no false statement, breached no access control, and forged no signature. They bought a token that was openly for sale and voted with it, which is the token’s entire designed purpose. That does not make it moral, and prosecutors may yet find angles, wire-fraud theories or sanctions violations tied to the Tornado Cash funding, but it does make it slippery. The uncomfortable implication is that “the vote is the exploit” is not only a security problem; it is a legal vacuum.

Regulators are still deciding who even answers for this. In the United States, jurisdiction over DeFi is split and contested between the SEC and the CFTC, and the question of whether a governance token is a security remains unsettled. Protocols, meanwhile, are left to self-insure through design, since there is no regulator waiting to reverse a bad block; we costed out that compliance burden and its limits in our look at DeFi compliance in 2026. For now, the most reliable defense against a governance attack is not a lawyer. It is a quorum threshold.

Who Pays When the Vote Is the Weapon?

Term’s own history sharpens the question. After the April 2025 oracle incident, the team absorbed the shortfall and reimbursed users. A governance attack is a different animal. There is no bug to patch and no insurer of last resort; the funds left through the front door with a valid ticket. Term Labs moved fast to contain the damage, shutting the Meta Vaults for good and revoking their governance roles so the same trick cannot be repeated, but containment is not restitution.

The Yearn wrinkle matters for accountability, too. Because the failure lived in Term’s custom wrapper rather than Yearn’s audited vault code, responsibility sits squarely with Term, not with the infrastructure it built on. That is an increasingly common shape in DeFi: a protocol composes on top of a trusted base layer, adds its own logic at the edges, and discovers that the edges are where the risk concentrated. The base layer can be flawless and the product still fail, because the vulnerability was introduced by the integration, not inherited from the dependency.

For depositors, the practical lesson is bleak but clarifying: in a leaderless protocol, “the DAO will make it right” is a hope, not a guarantee. The only party with a firm obligation to protect your deposit is you, exercising the one power an attacker cannot vote away, the choice of where to put your money in the first place.

How to Read a Vault’s Governance Before You Deposit

You do not need to be a smart-contract auditor to spot the conditions that made Term vulnerable. Most of the red flags are visible in a protocol’s governance documentation and a block explorer. The instinct a careful player brings to a provably fair game, verify the mechanism before you trust the outcome, is exactly the instinct a depositor should bring to a vault’s governance. Before you deposit, work through the list below.

• Is the governance token thinly held or thinly traded? A small float or a shallow market means a majority is cheap to buy. Term’s token was both.
• Can one proposal move funds? Check whether a single vote can transfer the treasury, mint tokens or upgrade contracts, or whether it can only adjust parameters within safe bounds.
• Is there a timelock, and who can veto during it? A delay is worthless without an independent, empowered party watching. Ask who that party is and whether they have ever actually acted.
• Are votes snapshotted from a past block? If voting power is measured at proposal time, an attacker can buy in after the fact. Past-block snapshots close that door.
• What is typical turnout? Chronic low participation is the precondition every attacker relies on. If quorum passes on a few wallets, so can a takeover.
• Who audited the governance layer? An audit of the underlying vault code is not an audit of the custom wrapper around it. Term ran on audited Yearn code and still fell.

The State of Governance Security Heading Into 2027

The summer of 2026 will be remembered as the season the governance attack graduated from a novelty into a repeatable business. The tooling to prevent it exists, snapshot voting, guardian councils, execution delays with teeth, concentration monitoring, and the more mature protocols have adopted much of it. Binance said its security team headed off a malicious proposal against an unnamed DAO’s roughly $1.2 million treasury with under 48 hours to spare, coordinating with other exchanges to freeze deposits until the project could reject it, though the exchange declined to name the target, so the claim cannot be independently verified. It is at least a sign that active defense can work when someone is genuinely watching. The gap is not knowledge. It is adoption, and it is widest at exactly the small and mid-sized protocols that make the most attractive targets.

Term Finance is the cautionary tale precisely because it was not careless. It had thought about governance risk, built in delays and vetoes and role separation, and still handed an attacker 68% of its vaults for a few thousand dollars. If a protocol that tried can lose this way, the ones that have not thought about it at all are simply waiting their turn. The uncomfortable takeaway for 2027 is that in decentralized finance, “trustless” was always a claim about the code. The votes still run on trust, and trust, as Term learned at 06:25 on a Sunday morning, is exactly what an attacker with two ETH and some patience can buy.

Frequently Asked Questions

What is a governance attack in DeFi?

A governance attack is when someone acquires enough of a protocol’s voting power, usually by buying a thinly held governance token, to pass proposals that hand themselves control of its funds. Unlike a code exploit, nothing is hacked in the traditional sense; the smart contracts execute exactly as written. The Term Finance attack drained about $8.5 million this way, with the attacker seeding the whole operation with just two ETH.

How did the Term Finance attacker steal $8.5 million with only 2 ETH?

Term’s vault governance token was sparsely held, so a small purchase bought a large share of the vote. Seeded by two ETH routed through Tornado Cash, the attacker reached up to 100% of the votes in four of five USDC vaults and about 91% of the Ethereum Meta Vault, then passed proposals that redirected 2,843 ETH and 1.68 million USDC to their own address. The assets those votes controlled were worth far more than the votes cost to acquire.

Did Term Finance have a timelock, and why did it not stop the attack?

It did: a seven-day timelock plus veto rights for liquidity providers. But a timelock only protects you if someone with the power to act is watching during the delay. Once the attacker controlled almost all of the voting power, they were effectively the only party with standing to veto, so the waiting period changed nothing. As Vitalik Buterin has put it, timelocks are more like a paywall on a newspaper than a lock and key.

Was the Yearn V3 code hacked in the Term Finance exploit?

No. The vaults ran on Yearn V3, but Yearn said the exploit came through Term’s custom governance wrapper and does not apply to standard Yearn vault setups. PeckShield and CertiK both concluded the attack targeted the voting mechanics, not any flaw in the underlying smart-contract code.

Is a governance attack illegal?

It is legally murky. In May 2025, a US judge vacated the fraud and manipulation convictions of Mango Markets exploiter Avraham Eisenberg, partly because the protocol was permissionless and executed automatically. When an attacker uses a governance system exactly as it was built, prosecutors struggle to show a law was broken, which is one reason these attacks keep happening. Jurisdiction over DeFi between the SEC and the CFTC also remains contested.

By Anneke de Vries, Security & Exploits Desk, HOGE Wire.