A DAO promises that a community can hold a treasury and steer a protocol without trusting a central operator. When that promise breaks, the failure is rarely the dramatic one people expect. The archetypal disaster – The DAO of 2016, drained of roughly 3.6 million ETH through a reentrancy bug and salvaged only by a contentious Ethereum hard fork – was a smart-contract failure. The deaths that followed are mostly governance and treasury failures: the code worked exactly as written, and that was the problem. This page is a neutral anatomy of how DAOs actually die, each mode illustrated by a documented case, so the structural risks can be told apart from the merely unlucky ones. For the narrower taxonomy of deliberate exploits, see governance attacks.
Most DAOs do not die loudly at all. They go quiet – proposals stop, contributors drift away, and a treasury sits under a multisig that a handful of people still technically control. The louder failures below are worth studying precisely because they expose, in fast-forward, the same weaknesses the quiet ones succumb to slowly.
The purest structural failure is when someone simply buys or rents the votes to pass whatever they like. If voting weight is read live from token balances, an attacker never has to own the tokens – they can borrow them for the length of a single transaction.
The definitive case is the Beanstalk attack of April 2022. An attacker took roughly a billion dollars in flash loans, converted them into the protocol's governance stake within one transaction, instantly commanded a supermajority, and passed a proposal that drained about $182 million from the treasury – all before the loan had to be repaid (CoinDesk). The lesson every serious DAO took from it: voting weight must never be readable by capital that is only briefly present – whether by snapshotting balances at a past block or by weighting on something transient capital cannot acquire in time.
Capture does not require a flash loan when quorum is thin enough. In February 2022 an attacker executed a "hostile governance takeover" of Build Finance DAO by quietly accumulating governance tokens, then passing a proposal – one the community's own Discord bot failed to surface – that handed them the token contract's minting keys and the treasury. They minted 1,107,600 BUILD and drained the liquidity pools, netting around $470,000; the DAO was left with no control over any of its own infrastructure (The Block). No exploit was needed – just an on-chain governance process operating exactly as designed while almost nobody was watching.
A subtler variant turns the DAO's own legitimacy against it. In October 2022, Avraham Eisenberg manipulated the price of the MNGO token on Mango Markets to borrow roughly $110 million against inflated collateral, then used the governance tokens he had acquired to vote through a proposal letting him keep about $47 million and shielding him from prosecution – later arguing the on-chain vote made it "legal" (Cointelegraph). A US jury disagreed and convicted him of fraud and market manipulation (The Block), and in May 2025 the trial judge overturned the convictions on venue and on the absence of any false statement (TRM Labs). The episode is a warning that a governance vote is not the same thing as legitimacy: a captured process can launder an attack into an "approved" decision.
Capture is cheap because participation is low. Empirical studies of on-chain governance repeatedly find that only a small fraction of eligible holders ever vote – often under 10%, and on some Snapshot spaces below 5% (Blockchain Governance: An Empirical Analysis of User Engagement on DAOs).
Layer on the fact that a large share of any token's supply sits inert in cold wallets, lost keys, or exchanges, and a DAO's effective quorum shrinks to a sliver of the nominal supply – a "phantom quorum" in which a holder of a couple of percent of the token can carry a proposal unopposed (Optimism writes the correction straight into its quorum rule, setting the threshold against active votable supply, where delegated power counts only if it has been used in the past six months: Optimism Operating Manual). Apathy is not merely a health metric; it is the attack surface that makes the takeovers above affordable. For the full anatomy of low participation – why holders rationally abstain, the empirical turnout figures, and the mitigation ladder – see Voter apathy and governance participation.
Paying people to participate is not a full fix either. Jupiter rewards JUP holders who actually vote with quarterly Active Staking Rewards, yet in June 2025 it nonetheless paused DAO voting entirely, citing a “breakdown in trust” – a reminder that subsidized turnout is not the same as durable legitimacy.
Even absent an attacker, capital-weighted voting (see voting mechanisms) tends to concentrate power over time rather than distribute it. Vitalik Buterin's Moving beyond coin voting governance lays out the core defect: a voter who merely borrows tokens has no lasting exposure to the outcome, so coin voting measures capital-at-hand rather than commitment-to-the-project. The empirical arc bends the same way – a 2026 review of decades of governance research found that DAOs keep re-centralizing, with active control collapsing back onto a small set of whales, delegates, and core contributors. A DAO can be perfectly decentralized on paper and a de-facto oligarchy in practice; the gap between the two is where trust quietly erodes.
The sharpest illustration is Aragon DAO – because it was executed deliberately, by the industry's own governance vendor. Aragon's token holders voted in June 2022 to move a nine-figure treasury on-chain; the Swiss association holding it delayed, and when holders assembled a majority to claim it in May 2023, the association halted the transfer and repurposed the DAO, describing those holders as raiders. It later dissolved and returned the treasury directly, concluding in its own words that "neither the AA nor ANT are currently suited to govern the project". Re-centralization here was not drift. It was a decision taken the moment decentralization became expensive, which is the only moment it is ever tested.
The treasury is the thing a DAO exists to steward, and it fails in two characteristic ways. The first is custody risk: the funds are, in practice, controlled by people the community cannot vet. Wonderland (the TIME project) ran a treasury that peaked above $1 billion, until its pseudonymous treasury manager "0xSifu" was revealed to be Michael Patryn, a convicted felon and co-founder of the collapsed QuadrigaCX exchange; assets under management cratered toward $146 million and the project unwound even after the community voted overwhelmingly to remove him (CoinDesk; The Block). The second is concentration risk: treasuries held largely in the DAO's own governance token show enormous paper valuations that evaporate the moment the token is actually sold, leaving little to fund the work the DAO promised.
Not every shortfall is terminal: when THORChain's ThorFi lending products left the protocol with more than $200 million of RUNE-denominated debt, its node operators paused the network and restructured the hole into a revenue-share token (THORChain) rather than inflating it away – a reminder that whether an insolvency ends a DAO turns as much on its governance as on its balance sheet. Sound treasury management – diversification, verifiable on-chain custody, and spending tied to enforceable rules – is what separates a durable DAO from a balance sheet waiting to be marked down.
The most radical answer is to have no treasury at all: Protocol Guild routes $183m of lifetime donations through immutable vesting contracts that, in its own words, "cannot be stopped or otherwise redirected during the vest by anyone, be it the donor or Protocol Guild membership" – neither failure mode can occur where no vote and no custodian can reach the funds. That escape is only available to an organization with nothing to decide, which is precisely what most DAOs are not.
The most common ending is the least reported. Contributors move on, proposal flow dries up, and the DAO becomes a dormant contract with a treasury nobody is actively directing. A large part of what drives this is the absence of a clean way out: when a member disagrees with the direction but cannot leave with their fair share of the treasury, the rational move is to disengage rather than fight – which drains exactly the engaged minority a DAO needs to stay alive. This is why exit rights (the "rage-quit" pattern popularized by Moloch DAOs) matter structurally: a credible exit keeps insiders honest and gives dissenters an alternative to quiet abandonment.
A DAO with no exit does not resolve its disagreements – it accumulates them until it stops. What that stopping actually looks like on-chain, and how it differs from a DAO that deliberately votes itself closed, is the subject of winding down a DAO. Not every ending is quiet, though: in June 2025 the ApeCoin DAO took the opposite exit, voting by a reported 99.66% to dissolve itself entirely and transfer its treasury and intellectual property to a founder-controlled company – a formal, near-unanimous surrender of the DAO form rather than a slow fade.
Caper is designed around these failure modes rather than the happy path. Decisions use ranked-choice voting over proposals that carry real on-chain actions, and voting weight is the product of two things: the governance stake a member holds and the participation they have actually shown – the latter earned as non-transferable tokens minted 1 per ballot cast and on nothing else. Because those tokens never leave the account that earned them, capital borrowed for the length of a block carries the lender's record rather than acquiring one, and the flash-loan route that drained Beanstalk simply has nothing to grip.
That structurally answers both the rented-majority capture and the phantom quorum, since influence accrues to members who have transacted with the caper themselves rather than to a position passing through someone else's hands. And because that same earned standing, not a raw token balance, sizes each member's claim on the treasury at exit, voice and exit are the same number: a member who disagrees can leave with their fair share rather than quietly abandon the DAO, and the treasury math is enforced by contract rather than entrusted to a custodian. The specifics live on the linked governance pages; this article is the map of the failures they are built to avoid.
Two Caper essays pick up where this map ends. They argue the deepest failures above trace to two missing structural rights – an owned treasury and an enforceable way out:
Related on this wiki: Rage-quit and exit rights · Personal tokens · DAO tokenomics.