← Back to Blog

Web3 Security · 6 min read · October 23, 2026

What Is a DAO Governance Attack? How a Vote Can Drain a Treasury in 2026

A DAO governance attack uses the protocol’s own vote to pass a proposal that hands over the treasury or the admin rights behind it. Nothing is hacked, no code fails, and an audit has nothing to find — which is why five incidents in 2026 still cost about $25.1 million.

What a Governance Attack Actually Is

A governance attack is an attempt to pass a malicious proposal through a DAO’s own voting process and use it to seize the treasury or the admin rights behind it. No vulnerability is exploited. No access check is missing. The contract authorizes the transfer because, as far as it can tell, the community simply changed its mind.

That is why these attacks survive audits: there is no defect to find. The only defence is someone reading the proposal before it executes — and in most 2026 cases, nobody did.

Three settings define the attack surface: the proposal threshold, the quorum and the timelock. Attackers read them as a price list.

The 2026 Scorecard

DefiLlama classified five incidents in 2026 as governance attacks, worth roughly $25.1 million combined. The headline loss was a $20.5 million malicious proposal against Bonk DAO in July. Term Finance lost about $8.5 million in August from its ETH meta vault. Yam Finance lost admin control of its timelock outright.

The stopped attempts are as instructive: Moonwell, Reserve, Olas and Ampleforth all had hostile proposals filed in 2026, each caught and cancelled before execution. The difference is not better code — it is whether a human was watching the queue.

The cheapest near-miss

In March 2026 an attacker spent about $1,808 to buy just over 40 million MFAM, self-delegated, and cleared Moonwell’s 40-million quorum with 0.4% to spare. Eleven minutes later they filed a “Protocol Recovery: Admin Migration” proposal that would have handed admin control of seven lending markets to an address they owned. Roughly $1.08 million of user funds was exposed — about $1 risked for every $600 takeable.

Five Ways Attackers Build Voting Power

• Self-delegation. In Compound-style governance, holding tokens is not enough; you must delegate. A wallet moving from zero voting power to above quorum in one transaction is the loudest signal in the system.
• Staked or vault shares. No delegation event fires at all, so the only signal is the staking contract’s own accounting. Term Finance’s ETH meta vault had so little staked that roughly half an ETH bought about 91% of voting power.
• Buying quorum outright. The Bonk DAO attacker spent about $4.09 million for 1.003% of supply — just over the approval threshold — recovered the stake after the vote and walked with $20.5 million.
• Flash loans. Beanstalk borrowed about $1 billion from Aave in April 2022, reached a two-thirds supermajority and executed in the same transaction, draining $182 million. Instant vote counting made it possible.
• Buying only the right to propose. The cheapest play, and the easiest to miss. In September 2026 an Olas proposer locked 5,100 OLAS — about $135 — clearing the 5,000 threshold by 2%. They never needed to pass anything themselves; they needed delegates to wave through something that looked routine.

How to Read a Malicious Proposal

Four things give a hostile proposal away, and all four are visible before it executes.

The payload rarely moves money directly. Watch the calldata for admin-surface functions: setPendingAdmin, changeOwner, transferOwnership, upgradeTo, grantRole. Moonwell’s targeted admin migration, Yam’s hit setPendingAdmin on the timelock, Olas’ targeted treasury ownership.

The description contradicts the calldata. Yam’s proposal #45 shipped with an empty description of literally 0x. Olas titled theirs “Owner migration: transfer treasury ownership from old timelock to Safe updater” — but the address named as the Safe had no code deployed. It was an ordinary wallet.

Legitimacy is cheap to fake. The Olas proposer registered autonolas-deployer.eth four days before proposing, pointed at their own address. A few dollars bought the appearance of being the protocol’s deployer.

The target contract changed. Reserve’s attempt pointed an upgrade at two core contracts set to unverified code. An upgrade to code nobody has reviewed is an admin transfer wearing a maintenance ticket.

Why a Timelock Is Not Enough

Term Finance had a seven-day timelock, and liquidity providers held veto rights over queued transactions. It still lost about 68% of TVL — roughly 2,843 ETH and $1.68 million USDC.

The malicious payload sat in plain sight for six days. It did two things at once: queued a strategy to drain the vault, and queued an instruction to zero out the seven-day cooldown itself. Nobody used the veto. The proposal executed twelve seconds after the voting window closed, disabling the safeguard in the same transaction.

A timelock buys time. It does not buy judgment, and it cannot protect itself if the ordinary proposal path can rewrite it. TOP Protocol in June 2026 had no timelock at all: an attacker funded with roughly 664 ETH from Tornado Cash bought a majority, minted new tokens and sold them for about $1.6 million.

The Four Layers That Actually Stop It

1. Time-weighted voting power. Average or lock voting power over a meaningful period so borrowed or freshly bought tokens cannot vote. This alone defeats the Term Finance and Beanstalk patterns.
2. A mandatory delay scaled to severity. Hours for routine parameters, days for treasury movements — long enough for someone to read the calldata.
3. A real guardian, held separately. Veto authority in the hands of a party outside the voting body, with an on-call path and monitored alerts. Compound added exactly this in proposal 304, a Proposal Guardian able to veto a passed proposal awaiting execution.
4. Scope restriction on governance itself. Governance must not be able to change its own timelock, quorum or guardian through the ordinary proposal path. This is the layer Term Finance was missing.

What Teams and Token Holders Should Do

If you run a DAO: alert on any delegation event that crosses quorum, simulate every proposal’s calldata before execution, and treat an upgrade to unverified code as an incident. If you hold governance tokens: delegate deliberately rather than by default, read proposals that touch admin functions, and treat a Friday proposal from an unfamiliar proposer as the most likely attack of the quarter.