What a DAO Is and Why Almost None of Them Are Truly Autonomous

Key points

A DAO does not run itself: an approved proposal is queued into a timelock contract by a person and executed by a transaction whose gas a human pays.
Abstention is the rule, not the exception. In most large DAOs quorum is reached thanks to a handful of delegates with aggregated voting power, not thanks to ordinary holders.
Almost every significant DAO keeps a multisig of between three and nine people able to pause contracts or move funds in an emergency. That is where the real power sits.
The flash-loan governance attack destroyed Beanstalk on 17 April 2022, with 182 million dollars in losses. The flaw was not the loan: it was having no delay between approval and execution.
Without a legal wrapper, a DAO can be treated as an unincorporated association, exposing its members to personal liability. There are already US rulings along those lines.

The acronym promises three things: decentralised, autonomous, organisation. The third one usually holds. The first holds partially. The second is, in the overwhelming majority of cases, false: practically no DAO runs itself. People draft the proposal, people queue it into the contract, people pay the gas for the transaction that executes it, and very often a small group holds an emergency key that can bypass the whole process.

That does not make DAOs a fraud. It makes them something more modest and more interesting: a mechanism for coordinating decisions about a shared treasury under rules that enforce themselves once approved. It is worth understanding how that mechanism works from the inside, because that is where both its strengths and its breaking points live.

The ideal versus the practice

The canonical definition is an organisation whose operating rules are encoded in smart contracts, whose decisions are taken by member vote, and whose execution requires no trusted intermediaries.

All three claims degrade on contact with reality:

Encoded rules. Only some of them are. Budgets, hiring contributors and legal strategy live in documents and spreadsheets, exactly as in any other organisation.
Decisions by vote. Formally, yes. Materially, voting power concentrates in a few addresses and participation is low.
No trusted intermediaries. This is where the model breaks: the emergency multisig is a trusted intermediary with names attached.

How a real vote works, step by step

The standard circuit on Ethereum has six phases, and each one has its own weak point.

Forum discussion. Somebody posts an idea, it gets debated and refined. No binding effect.
Off-chain signalling. A vote on a platform such as Snapshot, where signing is free because nothing is written to the chain. It measures sentiment; it cannot execute anything.
On-chain proposal. The exact set of calls to be executed — which contract, which function, which parameters — is submitted to the governance smart contract. Submitting it requires clearing the proposal threshold, a minimum amount of delegated voting power that keeps out spam.
Voting delay. A gap between the proposal being registered and voting opening, so everyone has time to read it.
Voting. Votes for, against and abstentions are counted. The proposal must clear quorum — a minimum number of votes cast — and the approval threshold, normally a simple majority of votes in favour over votes against.
Timelock and execution. The approved proposal is queued in a timelock contract that holds it for a fixed period. Once that expires, anyone can call the execution function. And that "anyone" is the crux: if nobody pays the gas for that transaction, the approved proposal simply does not happen. In practice it is always the same handful of team members or active delegates who do it.

Autonomous does not mean automatic. A proposal approved unanimously stays inert if nobody calls the contract to execute it. The chain guarantees that the instruction cannot be altered, not that anyone will carry it out.

Where the power really sits

Low turnout, concentrated votes

These are two opposite problems that feed each other. Mass abstention is the norm: most holders of a governance token never vote, because the cost of getting informed far exceeds the marginal influence of their vote. It is precisely the rational apathy of the minority shareholder, made worse by the fact that many hold the token to speculate, not to govern.

The result is that quorum is only reached thanks to a few heavyweight participants: investment funds that took part in private rounds, the project's own foundation, and professional delegates who have accumulated the voting power of thousands of small holders. In many large DAOs, somewhere between five and fifteen addresses decide any vote in practice.

Delegation is the patch for abstention: you hand your voting power to an address that votes on your behalf without you surrendering the tokens, and you can revoke it whenever you want. It keeps the system from seizing up, but it sharpens the concentration. Honesty requires putting it plainly: delegation does not democratise governance, it professionalises it.

The emergency multisig

Almost every significant DAO retains a multisig — typically 3-of-5, 4-of-7 or 5-of-9 signers — with one or more of these powers:

Pausing contracts during an ongoing attack, without waiting days for a vote.
Running the operating treasury: payroll, suppliers, grants already approved.
Upgrading the proxy contract that holds the protocol's logic.

The first two are defensible: no system can respond to an active exploit through a seven-day process. The third is the one worth examining closely, because it means that group can change the code holding user funds. Before trusting a protocol, find out who the signers are, how many signatures are required and whether that multisig is itself subject to a timelock. It is the same check I recommend in the DeFi guide, and for the same reason: it determines who can walk away with your money.

Governance mechanisms and their weak points

Mechanism	What it is for	Weak point
Proposal threshold	Preventing proposal spam	Excludes small holders; forces them to court a whale
Quorum	Stopping a minority deciding for everyone	Set too high it paralyses the org; set too low, capture is cheap
Token-weighted voting	Aligning votes with economic exposure	Whoever holds most tokens decides most. Buying power is legal and direct
Delegation	Countering abstention	Concentrates power in a few professional delegates
Off-chain voting (Snapshot)	Voting without paying gas	Executes nothing by itself; depends on somebody carrying it on-chain
Timelock	Creating room to react to malicious proposals	Also delays legitimate emergency responses
Emergency multisig	Reacting fast to an attack	A central point of trust and of failure
Voting-power snapshot	Preventing vote buying mid-ballot	If the snapshot is taken late or badly, it can be gamed

When governance fails

The founding case: The DAO, 2016

The DAO was a collective investment fund deployed on Ethereum in 2016. It raised more than 12 million ETH from thousands of participants, a very substantial share of all ETH in circulation at the time. Its contract let anyone who disagreed with a decision "split off", withdrawing their share into a child contract.

On 17 June 2016 an attacker exploited that split function through a recursive call: the logic sent the funds before updating the internal balance, so the same withdrawal could be repeated many times inside a single transaction. This is the bug now known as reentrancy. According to the Ethereum Foundation's own record, more than 3.6 million ETH were drained. The funds ended up locked in a child contract subject to a waiting period, which bought time to decide what to do.

The decision was a hard fork: altering the chain's state to return the funds to their contributors. It activated at block 1,920,000 on 20 July 2016. Some miners and users rejected it, arguing that the flaw was in the contract and not the protocol, and carried on producing blocks on the original chain. That chain is Ethereum Classic (ETC); the one that applied the fork is what we now call Ethereum (ETH).

Both lessons are uncomfortable and still current. First: "code is law" lasted exactly until the amount at stake got large enough. Second: the ability to reverse things existed because a small group of people could coordinate the miners. That is not autonomy; it is social governance operating above the code.

Governance attacks

If voting power can be bought or rented, governance can be bought or rented. There are two variants.

Patient accumulation: somebody buys governance tokens on the open market until they can pass proposals. It is slow, expensive and visible, but perfectly legal, and it has happened at protocols whose treasuries exceeded their own market capitalisation, where buying control costs less than the prize.

Instant capture with a flash loan: the attacker borrows an enormous quantity of voting tokens within a single block, uses it to approve their own proposal and executes it before repaying the loan, all inside the same transaction. That is what happened to Beanstalk on 17 April 2022, with losses of 182 million dollars. The design error was not allowing flash loans — an ecosystem primitive that cannot be banned — but allowing a proposal to be voted on and executed with no delay in between.

The standard defences are three: take a snapshot of voting power at a block before the proposal, so that buying tokens afterwards achieves nothing; impose a voting delay so nobody can vote in the same block a proposal is made; and add a timelock before execution. A protocol with all three is very hard to capture in one block. One without them is an invitation.

Worked example: a proposal's full journey

Here I follow the parameters published by Compound for its Governor Bravo system, which served as the template for much of the sector. Every DAO sets its own, so always check the specific protocol's.

Suppose the proposal is to raise an asset's collateral ratio from 75 % to 80 %.

Phase	Duration	What happens exactly
Forum discussion	5-7 days	The proposal is published with its risk rationale. No binding effect
Snapshot signalling	3-5 days	Signature-based vote, no gas cost. Gauges support before spending
On-chain submission	One transaction	The proposer needs 25,000 COMP of delegated voting power
Review period	~2 days	The proposal is registered but voting has not opened
Voting	~3 days	At least 400,000 votes in favour must be cast for it to pass
Timelock	~2 days	The approved instruction waits in the queue; anyone can read it and react
Execution	One transaction	Somebody calls the execution function and pays the gas. The parameter changes

Total: between 15 and 19 days from idea to the number changing in the contract. That timetable is precisely why the emergency multisig exists: no protocol can afford two weeks to respond to an attack already under way.

Note one further detail in step three. A small holder with 50 COMP cannot propose anything alone: they need to persuade a delegate who aggregates 25,000. Open governance is open for voting, far less so for proposing.

The legal problem: who is answerable

This is the question the sector finds most uncomfortable. If a DAO causes harm — a badly parameterised protocol that liquidates users improperly, an operation that breaches a country's rules — who gets sued?

The default answer in several legal systems is the worst possible one for participants: if a collective acts in common for profit without having incorporated a legal entity, it can be treated as an unincorporated association or a de facto partnership, with unlimited joint and several liability for its members. There is already precedent in the United States: in the Ooki DAO matter, the CFTC obtained a default judgment in June 2023 in which the court accepted treating the DAO as an unincorporated association capable of being sued and sanctioned. The implication is direct: token holders who voted could be considered members.

Hence the proliferation of legal wrappers:

Wyoming DAO LLC. A limited liability company with a DAO-specific regime, in force since 2021, granting legal personality and limiting member liability. Wyoming later added an unincorporated non-profit association form aimed at non-profit DAOs.
Swiss foundation or association. The classic wrapper for large projects, with assets dedicated to a purpose and identifiable governing bodies.
Cayman Islands foundation. A foundation company with no shareholders, used to hold the treasury and contract on the project's behalf.
Nothing at all. Still the most common choice for small DAOs, and the one with the greatest personal risk for participants.

In the European Union the obstacle is structural. MiCA conditions authorisation as a crypto-asset service provider on the applicant being a legal person with a registered office in a member state and identifiable effective management. A DAO with no legal personality cannot, by definition, apply for that authorisation: either it adopts a corporate form, or it stays outside the regulated perimeter with everything that entails. The detail on how supervision and timelines are split is in the crypto regulation guide.

What they are genuinely good for today

Strip out the enthusiasm and three uses with demonstrable results remain:

Protocol treasury management. Deciding what happens to a protocol's revenue through a public, auditable procedure. It is dull and it works.
Grant programmes. Distributing funds to developers and ecosystem projects, with voting and on-chain accountability.
Risk parameters. Adjusting collateral ratios, interest rates and accepted assets in lending protocols. Here the DAO acts as an open risk committee, and its value is that decisions and the reasoning behind them are on the record.

What they have not demonstrated is the ability to replace a company in decisions requiring speed, confidentiality or identifiable personal responsibility.

What to review before taking part

Find the governance contract and note three numbers: proposal threshold, quorum and timelock duration.
Check whether a multisig exists with the power to upgrade contracts, how many signatures it needs and who the signers are.
Look at actual turnout across the last ten votes, not the theoretical figure. Count how many addresses decided the outcome.
Verify that voting power is calculated from a snapshot taken before the proposal.
Find out whether the DAO has a legal wrapper and which one. If it has none, weigh your personal exposure before taking a visible role.
Read the exact text of the calls a proposal executes, not just its forum summary. What runs is the bytecode, not the explanatory note.
If you delegate, check periodically how your delegate has voted and revoke if they no longer represent you.

A well-built DAO is a system of transparent rules with its human control points declared. A badly built one is an opaque company with a decorative layer of voting on top. You can tell the difference in five minutes by reading its governance contract and its multisig, and you cannot tell it at all by reading its landing page.

Frequently asked questions

What is the difference between a DAO and a company that holds votes?

Less than the name suggests. The real difference is that in a DAO the outcome of a vote can execute directly against a contract holding funds, with nobody having to sign a document. Everything else — who proposes, who holds votes, who executes in practice, who can halt the system — looks a lot like a corporate governance body, sometimes with fewer safeguards.

Can I lose money by taking part in a DAO?

You can lose the value of the governance tokens and, in extreme scenarios, be exposed to legal liability. If the DAO has no legal personality and a court treats it as an unincorporated association or a de facto partnership, active members could be answerable for its obligations. It is an unlikely but real risk, and it depends on the applicable jurisdiction.

What is a timelock and why does it matter so much?

It is a contract that receives an approved instruction and holds it for a fixed period — usually between one and seven days — before allowing execution. Its purpose is not bureaucratic: it creates room for somebody to spot a malicious proposal and for anyone who disagrees to withdraw their funds. Governance without a timelock can be attacked inside a single block.

Is delegating my vote worth it?

It is worth it so that your weight is not lost to abstention, but it concentrates power. A delegate with large aggregated voting power can meet the proposal threshold alone and decide the outcome of almost any vote. If you delegate, review your delegate's voting record periodically: delegation is usually revocable at any time.

Sources and references

Important notice. InfoCrypto is an educational site. Nothing here is financial, tax or investment advice. Crypto assets are high risk: you can lose all the capital you invest. Always verify data against the original sources before making any decision, and consult a licensed professional if you need to.