DAO governance in practice: what the on-chain evidence shows

Decentralised autonomous organisations (DAOs) promise governance without managers: token holders propose, deliberate and vote, and smart contracts execute the result. Since 2020 a body of empirical work has tested that promise against on-chain records. This study reviews the main quantitative studies of DAO voting (on participation, token concentration and delegation), the documented governance attacks, the legal wrappers that states have created for DAOs, and one notable attempt to correct token plutocracy by institutional design, the Optimism Collective's two-chamber system. The evidence is consistent and unflattering: in most large DAOs a handful of addresses can decide any vote, most tokens do not vote, delegation tends to concentrate rather than spread power, and governance can be bought, sometimes within a single transaction. The evidence also has limits. On-chain data show addresses, not people; they show votes, not the off-chain bargaining that precedes them; and most studies cover a few large Ethereum protocols in 2020-2023. The lessons for institution design are nonetheless clear enough to state: token-weighted voting reproduces shareholder governance without its legal protections, and the most promising reforms either change what counts as a vote (one person, one vote; non-transferable membership) or add time and checks (execution delays, vetoes, second chambers).

Key findings

• Voting power is highly concentrated. In Compound, Uniswap and ENS, eight, eleven and eighteen delegates respectively controlled more than half of delegated voting power in March 2022, with Gini coefficients among delegates of 0.987, 0.995 and 0.908 (Fritsch, Müller and Wattenhofer). Across 21 DAOs, fewer than ten participants controlled a majority in 17 (Feichtinger and colleagues).
• Participation is low. In the Compound and Uniswap data, on average less than 10% of total token supply (about 15% of circulating supply) took part in votes, and 7.5% of circulating supply would have sufficed to win an average vote.
• Delegation concentrates power more often than it spreads it. In Compound and Uniswap, delegates who represent many small holders held under 10% of delegated votes; most voting power sat with delegates acting for a single holder. A 2026 study of 48 Ethereum DAOs argues that delegation, staking and token registration systematically reinforce centralisation.
• Governance can be bought. The Beanstalk attack of April 2022 used flash-loaned tokens to pass and execute a malicious proposal in a single transaction, draining about $181 million from the protocol; the attacker kept about $76 million. Other cases (Build Finance in 2022, a contested Compound treasury proposal in 2024) show slower forms of capture through token accumulation.
• Insiders often decide. Contributors to a project held, on average, the majority needed to control decisions in at least 7.5% of DAOs studied and single-handedly decided at least one proposal in about one in five; token holdings shifted shortly before the vote in about 15% of 8,116 proposals (Kitzler and colleagues).
• Legal wrappers now exist, but liability risk remains. Wyoming (DAO LLC, 2021; DUNA, 2024) and the Marshall Islands (DAO LLC) offer legal personality and limited liability. Without such a wrapper, US courts and regulators have treated DAOs as persons that can be sued and penalised, and have allowed claims that members are general partners.
• Design can change outcomes, but evidence is thin. One-person-one-vote DAOs became more decentralised over time in one comparative study, and Optimism's bicameral model separates token-weighted from citizen-based decisions. Neither has yet been evaluated with the rigour applied to token voting.

Research question and scope

The question is descriptive and institutional: how do DAOs actually govern themselves, and what does the evidence imply for the design of digital collective decision-making? "DAO" here means an organisation whose collective decisions are taken by voting with blockchain-based tokens or credentials, whether the vote is recorded on-chain (with automatic execution) or off-chain (for example on the Snapshot platform, with execution by a multisignature wallet).

The study covers five topics: (1) participation; (2) the distribution of voting power; (3) delegation; (4) governance attacks and capture; and (5) legal form and liability. It closes with one case of deliberate institutional redesign, the Optimism Collective. It does not assess the economic performance of DAOs, their security apart from governance, or the merits of particular protocols. Prices and dollar values are those reported by the cited sources at the time and fluctuate heavily with token markets.

Background

The term entered wide use with "The DAO", an investment fund launched on Ethereum in 2016 that was drained through a smart-contract flaw within weeks; the Ethereum community's decision to reverse the theft through a hard fork remains the best-known example of off-chain politics overruling on-chain rules. The modern wave began in 2020 with decentralised finance (DeFi) protocols such as Compound and Uniswap, which distributed governance tokens to users and investors and handed parameter changes, treasury spending and upgrades to token votes.

Most of these systems share a template. A governance token confers one vote per token. Holders must delegate their votes (to themselves or to someone else) before voting; delegation saves transaction costs because a delegate votes once for everyone who delegated to them. A proposal needs a minimum token threshold to be submitted, a quorum to be valid, and a majority to pass. On-chain systems then queue the decision in a timelock before execution. Off-chain systems record votes for free and rely on a small group of signers to execute them.

From the perspective of political theory this is a hybrid: it borrows the vocabulary of direct democracy, the delegation mechanics of liquid democracy, and the weighting rule of a joint-stock company. The empirical question is which of these the practice most resembles.

Method and sources

This is a structured review of published evidence, not new data collection. Sources were selected for direct relevance and methodological transparency:

• Quantitative on-chain studies: Fritsch, Müller and Wattenhofer (data to 1 March 2022; journal version 2024); Feichtinger, Fritsch, Vonlanthen and Wattenhofer (21 DAOs, 2023); Barbereau and colleagues (nine DeFi projects in the Bloomberg Galaxy DeFi Index, 2023); Kitzler and colleagues (contributors and pre-vote token shifts across thousands of proposals, 2025); Pahari and colleagues (design analysis of 48 Ethereum DAOs, 2026).
• Mixed-methods work: Sharma and colleagues (ten DAOs, combining on-chain data, forum discussions and interviews, 2023).
• Incident records: the public post-mortem literature on Beanstalk, press reporting on Build Finance, and the on-chain record of Compound Proposal 289.
• Primary legal texts and decisions: Wyoming Enrolled Acts 73 (2021) and 23 (2024); documentation of the Marshall Islands DAO LLC regime; the CFTC's statement on its default judgment against Ooki DAO; the decision in Samuels v. Lido DAO.
• Constitutional documents: Optimism's Working Constitution (2022) and Operating Manual (version 2.0.0).

Most studies are preprints or conference papers in computer science, a field that publishes quickly; some have since appeared in journals. All on-chain studies share a basic limitation: an address is not a person. One person may control many addresses, and one address (an exchange, a custodian, a multisignature wallet) may represent many people. Concentration measured over addresses can therefore overstate or understate concentration over people.

Findings

Participation: most tokens do not vote

Fritsch, Müller and Wattenhofer analysed every delegation and vote in Compound, Uniswap and ENS from November 2019 to March 2022. Across votes, on average less than 10% of total token supply (roughly 15% of circulating supply) took part. They estimated that holding 7.5% of circulating supply would have been enough to win an average vote. For comparison they cite participation of around 70% among shareholders of US listed companies, where institutional investors are under legal and regulatory pressure to vote.

Low turnout is not a problem by itself; representative systems tolerate it. It becomes a problem when it combines with concentration, because it lowers the cost of control. Barbereau and colleagues, studying the nine projects in the inaugural Bloomberg Galaxy DeFi Index, found that voting rights were highly concentrated and that "the exercise of these rights is very low", and argued that "minority rule, not 'democracy', is the probable outcome" of tradable voting rights in the absence of anti-concentration rules. They call the result "timocratic", rule by property.

Feichtinger and colleagues add a less discussed cost. On Ethereum, governance transactions cost gas. Across 21 DAOs they found governance transaction costs of about $3.5 million for ENS (driven by delegations after its airdrop) and below $300,000 for each of the others; for Uniswap, the extra cost that the governance-enabled token contract imposed on ordinary transfers came to just under $3 million. They also found that most DAOs had more than 10% "pointless" votes and delegations (for example, votes cast by addresses with no voting power), many over 20%, which they read as a sign of immature systems. Off-chain voting removes transaction costs but shifts trust to whoever executes the result.

Concentration: a handful of addresses can decide

The Nakamoto coefficient, the smallest number of participants who together hold more than half of voting power, is the simplest summary of control.

Source: Fritsch, Müller and Wattenhofer, Tables 1-2. For comparison the authors cite 2020 wealth Gini coefficients of 0.850 for the United States and 0.814 for Europe.

Feichtinger and colleagues extended the analysis to 21 DAOs with on-chain governance. Except for four projects, all had single-digit Nakamoto coefficients among delegates; for half, the coefficient was no larger than three. In 17 of the 21 systems, fewer than ten participants controlled a majority of voting power, which "suffices to decide any vote". The authors observed a slight trend towards decentralisation over time, but inequality stayed high throughout.

Concentration in potential power does not always become concentration in exercised power. Fritsch and colleagues found that a single large Compound delegate could have changed the outcome of five of the nine proposals on which it voted, but that large delegates generally used far less power than they held; the second-largest Uniswap delegate, holding about 12.8 million UNI from a single address, had never voted. Large and small delegates also mostly voted the same way: they diverged on 6 of 84 Compound proposals and 1 of 10 Uniswap proposals. Concentration therefore looks less like constant domination than like a latent veto: it matters most in the close, contested votes where it is decisive.

Insiders, contributors and pre-vote trading

Kitzler, Balietti, Saggese, Haslhofer and Strohmaier examined whether a project's own contributors (developers, team members and other vested users) dominate governance. Across the DAOs in their data, contributors on average held the majority needed to control decisions in at least 7.54% of DAOs and alone decided at least one proposal in 20.41%. Contributors were centrally placed in the networks linking voters, which the authors read as evidence of inner power circles. They also found that governance-token ownership shifted shortly before the vote in 1,202 of 8,116 proposals (14.81%), consistent with tokens being acquired or borrowed to influence specific decisions.

Sharma and colleagues, combining on-chain data with forum analysis and interviews in ten DAOs, found poor voting decentralisation in Compound, AssangeDAO, Bankless and Krausehouse, and varying "autonomy": some DAOs relied heavily on third parties, such as a core company or a multisignature committee, to propose and execute decisions. The one-person-one-vote DAO in their sample, Proof of Humanity, became more decentralised over time. The finding is suggestive rather than conclusive (one case), but it points to the voting rule, not the technology, as the source of concentration.

These findings echo Robert Michels' "iron law of oligarchy": organisations that begin with broad participation tend to concentrate control in a small, active, well-resourced group. In DAOs the mechanism is unusually visible because the ledger is public.

Delegation: liquid democracy in theory, proxy concentration in practice

Delegation is meant to let passive holders lend their weight to informed, active representatives, a form of liquid democracy. The evidence shows how far practice diverges from that idea.

• Who delegates are. Fritsch and colleagues defined "community delegates" as those receiving less than half their votes from any single holder. In ENS, which required delegation as part of its airdrop, community delegates held about 60% of delegated voting rights. In Compound and Uniswap they held under 10%: most voting power sat with delegates who were, in effect, a large holder voting its own tokens.
• How much is delegated. In Uniswap only about 20% of total supply (about 30% of circulating supply) was delegated at all, so roughly 15% of circulating tokens could control governance.
• Design effects. Pahari, Chandrasekaran, Messias, Gummadi and Dash studied the designs of 48 actively used Ethereum DAOs with substantial capital and argue that three mechanisms introduced to improve security or participation (token registration, staking and delegation) instead contribute to concentrating voting power.

Delegation also reduces costs: Feichtinger and colleagues estimate that, without it, vote transactions would have cost far more. The trade-off is real. A system without delegation would see even lower participation; a system with delegation needs rules about who may be a delegate, how much one delegate may hold, and how delegations expire.

Governance attacks and capture

Beanstalk is the canonical case because it shows that, with flash loans, voting power need not be owned even for a minute. The attacker borrowed roughly a billion dollars' worth of stablecoins from lending protocols, converted them into governance-eligible positions, voted for a proposal submitted a day earlier, and executed it, all within one transaction, then repaid the loans. The post-mortem record attributes the attack to the absence of an execution delay: the emergency-commit path allowed a supermajority to execute immediately. The lesson is mechanical but general: any rule that measures voting power at the moment of voting, rather than at an earlier snapshot, and allows immediate execution can be bought with borrowed capital.

Build Finance shows the slower version. The Build Finance DAO's governance gave token holders control over the token contract. According to The Block's report of 14 February 2022, a first proposal that would have let its author mint tokens unilaterally was voted down; the actor then moved its tokens to a separate wallet, submitted a second proposal that the community's Discord bot did not flag, and passed it on 10 February. The proposal handed over the governance contract, minting keys and treasury. The attacker minted new BUILD tokens, drained liquidity pools on Balancer and Uniswap and sold treasury holdings, taking about 160 ETH (roughly $470,000). A core team member described a "total and irrecoverable loss" of treasury assets. These figures come from press reporting and the team's statements, not from peer-reviewed analysis.

Compound Proposal 289 illustrates a grey zone between attack and politics. The proposal, created on 24 July 2024 by a group of large holders calling itself the Golden Boys, sought a one-year investment of 499,000 COMP from the treasury into a vault (goldCOMP) whose multisignature wallet the group would control. It was legal under the protocol's rules and passed by about 682,000 votes to 634,000, above the 400,000 quorum; opponents called it a governance attack because it would have placed treasury assets under the control of a few large holders. It was cancelled before execution. Whether it was an attack depends on what one thinks a DAO's treasury is for, a question the code cannot answer.

Common defences have emerged: voting power measured at a snapshot block before the proposal, timelocks between approval and execution, guardians or security councils with power to veto or pause, and proposal thresholds. Each trades decentralisation against safety, and each creates a new locus of power that must itself be governed.

Legal wrappers and liability

A DAO without a legal form is, in many legal systems, likely to be treated as a general partnership or unincorporated association, with members potentially liable for the organisation's debts. Two kinds of response have appeared: statutes that create purpose-built forms, and enforcement actions that treat unwrapped DAOs as legal persons.

Wrappers solve some problems: they let a DAO sign contracts, pay taxes, hold property and shield members. CityDAO, an experiment documented in this atlas, organised as a Wyoming DAO LLC to buy land. Wrappers also import obligations: registered agents, filings, beneficial-ownership disclosure and, in the Marshall Islands, identity checks that sit uneasily with pseudonymity. The Ooki and Lido cases show the cost of not wrapping: regulators and plaintiffs can treat token voters as members of a partnership, so the people most exposed are identifiable, active voters. That creates a perverse incentive against participation, the opposite of what governance needs.

Institutional redesign: Optimism's two chambers

The Optimism Collective, which governs a family of Ethereum "layer 2" networks, is the most developed attempt to correct token plutocracy by constitutional design. Its Working Constitution, posted in April 2022, establishes "a bicameral system" in which OP token holders and "Citizens" coexist and check each other. Among its interpretive principles are "governance minimization", "forking and exit" and "anti-plutocracy". It describes itself as temporary, to expire no more than four years after adoption in favour of a permanent constitution.

• Token House. Token holders, voting directly or through delegates, decide protocol and governor upgrades, treasury appropriations, elections and certain removals. Under the Operating Manual (version 2.0.0), most proposals need explicit approval from four of the top 100 delegates (or four Citizens) to reach a vote; quorum is 30% of "votable supply", meaning delegated OP that has been used in a vote in the previous six months; approval thresholds are typically 51% or 76%. Many decisions are subject to a veto period.
• Citizens' House. Citizens are admitted by attestation rather than by buying tokens, and the Working Constitution gives them the allocation of retroactive public-goods funding, rewards paid after the fact for work judged to have benefited the ecosystem. The intent is that one chamber represents capital and the other represents contribution.
• Foundation. The Optimism Foundation, a Cayman Islands entity, stewards the system during the transition, with power to administer governance and amend the Working Constitution, which token holders can check by removing directors and vetoing harmful changes to founding documents.

As an institutional idea, this is close to the classical mixed constitution: different bases of representation in different chambers, with vetoes between them. What it has not yet produced is independent evidence that it works better. The selection of Citizens, the Foundation's continuing discretion and the veto rules are all young and frequently revised. It is best read as a live experiment, not a proven model.

What the evidence does not show

• That DAOs are uniquely oligarchic. Comparable data on member participation and control in cooperatives, associations or listed companies are rarely as complete. The low shareholder-style turnout in DAOs is real, but comparisons with institutional-investor-dominated corporate voting are imperfect.
• That concentration is always captured. Large delegates often vote with small ones and frequently do not vote at all. Concentrated potential power is a vulnerability, not proof of abuse.
• People behind addresses. None of the on-chain studies can identify individuals. Exchanges and custodians may hold tokens for many users; one actor may split holdings across addresses.
• Off-chain politics. Forum debates, private negotiation between large holders, and the influence of core development teams are mostly invisible to on-chain data. Sharma and colleagues' interviews show this layer matters.
• Generalisability. Most data come from a few large DeFi protocols on Ethereum between 2020 and 2023. Smaller, social or one-person-one-vote DAOs are under-studied.
• The effect of legal wrappers. There is no systematic evidence yet on whether wrapped DAOs govern better, last longer or face fewer disputes.

Implications for institution design

1. The voting rule matters more than the ledger. Token-weighted voting is shareholder voting. If the aim is democratic legitimacy rather than investor control, the unit of voting must be a person or a member, which requires some form of proof of personhood or non-transferable membership.
2. Measure power before the vote and execute after a delay. Snapshots taken before a proposal is published, timelocks between approval and execution and the absence of emergency fast-tracks defeat flash-loan attacks. They are cheap and should be default.
3. Treat delegation as representation. Delegates should have disclosed interests, caps or decay on delegated weight, and periodic re-confirmation, as representatives do in other institutions.
4. Add a second check that cannot be bought. A chamber, council or veto holder selected by a different principle (contribution, sortition, expertise) makes capture harder, provided its own selection is transparent and contestable.
5. Design for low participation. Quorums tied to active rather than total supply, and approval thresholds that rise with the stakes, keep systems workable without letting a small minority decide by default.
6. Settle the legal question early. An unwrapped DAO exposes its most active voters to partnership liability. A wrapper, or an explicit decision not to use one, should be part of the founding design.

Open questions

• Can proof-of-personhood systems support one-person-one-vote DAOs at scale without recreating centralised identity registries, and what do they cost in privacy and exclusion?
• Do bicameral or council-based models such as Optimism's measurably reduce capture, or do they move power to the bodies that select citizens and councillors?
• How should courts allocate liability in DAOs where a few delegates decide and thousands of holders never vote?
• Does participation rise when votes concern concrete, local decisions (as in CityDAO's land) rather than protocol parameters?
• What happens to DAO governance when AI agents, acting for holders, become delegates or voters?

Sources

1. Fritsch, R., Müller, M. and Wattenhofer, R. (2024). Analyzing voting power in decentralized governance: Who controls DAOs? Blockchain: Research and Applications 5(3), 100208. Preprint 2022: arXiv:2204.01176.
2. Feichtinger, R., Fritsch, R., Vonlanthen, Y. and Wattenhofer, R. (2023). The Hidden Shortcomings of (D)AOs: An Empirical Study of On-Chain Governance. arXiv:2302.12125.
3. Barbereau, T., Smethurst, R., Papageorgiou, O., Sedlmeir, J. and Fridgen, G. (2023). Decentralised Finance's timocratic governance: The distribution and exercise of tokenised voting rights. Technology in Society 73, 102251.
4. Kitzler, S., Balietti, S., Saggese, P., Haslhofer, B. and Strohmaier, M. (2025). The Governance of Decentralized Autonomous Organizations: A Study of Contributors' Influence, Networks, and Shifts in Voting Power. In Financial Cryptography and Data Security, Lecture Notes in Computer Science, pp. 313-330. Preprint: arXiv:2309.14232.
5. Sharma, T., Kwon, Y., Pongmala, K., Wang, H., Miller, A., Song, D. and Wang, Y. (2023). Unpacking How Decentralized Autonomous Organizations (DAOs) Work in Practice. arXiv:2304.09822.
6. Pahari, V., Chandrasekaran, B., Messias, J., Gummadi, K. P. and Dash, A. (2026). On the Centralization of Governance Power in Decentralized Autonomous Organizations. arXiv:2604.25959.
7. Rekt (2022). Beanstalk - REKT, 18 April 2022.
8. Tally. Compound Proposal 289: Trust Setup for DAO investment into GoldCOMP (on-chain proposal record, July 2024).
9. Wyoming Legislature (2021). Enrolled Act No. 73 (SF0038): Decentralized autonomous organizations. W.S. 17-31-101 to 17-31-116.
10. Wyoming Legislature (2024). Enrolled Act No. 23 (SF0050): Wyoming Decentralized Unincorporated Nonprofit Association Act. W.S. 17-32-101 to 17-32-129.
11. MIDAO. Marshall Islands DAO LLC documentation (service provider's summary of the Marshall Islands DAO legislation).
12. Commodity Futures Trading Commission (2023). Statement of CFTC Division of Enforcement Director Ian McGinley on the Ooki DAO Litigation Victory, 9 June 2023.
13. United States District Court for the Northern District of California (2024). Samuels v. Lido DAO, No. 23-cv-06492-VC, Order re Motions to Dismiss, 18 November 2024.
14. Optimism Collective (2022). Working Constitution of the Optimism Collective, 26 April 2022.
15. Optimism Collective. Operating Manual of the Optimism Collective, version 2.0.0.
16. Copeland, T. (2022). Build Finance DAO suffers 'hostile governance takeover,' loses $470,000. The Block, 14 February 2022.
17. Republic of the Marshall Islands (2022). Decentralized Autonomous Organization Act 2022, P.L. 2022-50 (PacLII).