Skip to content
Crypto.world

Research

On Exercising Governance Power in Decentralized Autonomous Organizations

DAOs let token holders govern blockchain protocols through rules encoded in a "governance contract," and anyone can buy governance tokens anonymously to acquire voting power.

On Exercising Governance Power in Decentralized Autonomous Organizations
arXiv
2607.26204
Published
2026-07-28
Authors
Vabuk Pahari, Balakrishnan Chandrasekaran, Johnnatan Messias, Krishna P. Gummadi, Abhisek Dash

AI summary

Overview

Research area: Blockchain governance — specifically the design of governance smart contracts in Ethereum-based Decentralized Autonomous Organizations (DAOs), with a focus on security (trust) and transparency trade-offs.

Technical level: Intermediate. Readers should know roughly how smart contracts, governance tokens, and on-chain voting work, but the paper's arguments are about mechanism design rather than cryptography or code-level exploits.

Scope: A systematic review of the governance contracts of 48 public, actively used Ethereum DAOs with large treasuries, mapping their design choices across the full lifecycle of a referendum and showing how those choices create a new class of "governance attacks."

What This Paper Is About

DAOs let token holders govern blockchain protocols through rules encoded in a "governance contract," and anyone can buy governance tokens anonymously to acquire voting power. To protect themselves, DAOs bolt on safeguards — proposal thresholds, delays, vetoers, multi-signature executors — but each safeguard trades away some decentralization, transparency, or security. The paper dissects how 48 real DAOs implement these mechanisms across proposing, voting, certifying, and executing a proposal, and shows that flawed design choices (not buggy code) leave many DAOs exploitable.

Key Contributions

  1. A structured review of the smart contract implementations and developer documentation of 48 public Ethereum DAOs, identifying the key admission controls ("gatekeeping") and "guard rails" used in their governance protocols (summarized in Table I).
  2. A lifecycle-based analysis of design choices — from the birth of a referendum, through voting, to certification, execution, and veto — with an explicit account of the trust and transparency trade-offs each choice implies.
  3. The identification of a new class of attacks the authors call governance attacks, which exploit the fundamental design of governance mechanisms even assuming bug-free implementations, illustrated with real-world case studies.
  4. A vulnerability assessment showing which of the 48 DAOs remain exposed to the design flaws exploited in recent attacks, plus the commitment to release the collected data and analysis code as open-source artifacts upon publication.

Main Findings

  • Two-thirds of the sample still allow vote sniping. Excluding only 4 DAOs (Lido, Frax, Origin, Angle), all studied DAOs use governance contracts that do not prevent vote sniping, where a voter withholds votes until the final moments to decide the outcome.

  • Vote sniping decided the Compound attack. In the July 2024 Compound governance attack, 563,591 of 682,191 votes (82% in favor) were cast within 170 blocks (34 minutes) of the end of voting, with the final vote 40 blocks (8 minutes) before the deadline. The referendum passed by fewer than 50,000 votes. Three malicious referendums were run in total; the first two failed, and voters likely assumed the third would fail too.

  • Governance power was assembled largely by purchase. 29 malicious wallet addresses bought tokens over the preceding 4 months: 563,790 tokens arrived via four centralized exchanges (292,570 through KuCoin, 230,335 through ByBit, 33,992 through HTX, 4,391 through OKX), and a further 118,089 tokens were borrowed through the Compound Protocol. Attackers held over 680,000 tokens, while forum discussion suspected roughly 325,000.

  • Fifteen DAOs have no oversight at all. 15 of the 48 DAOs have no oversight mechanism (such as a certifier or vetoer) to stop a malicious referendum, and 7 DAOs — Uniswap, Radicle, Gitcoin, Silo, Ampleforth, Hop, and Cryptex — combine all three flaws (buyable voting power, vulnerability to vote sniping, and no oversight).

  • 16 of 28 known DAO incidents are governance attacks. Building on Feichtinger et al.'s catalog of 28 incidents, the authors classify 16 as governance attacks, where a change in the governance mechanism would have prevented the attack. 6 of the 16 stemmed from a bug in the governance contract; 10 were token acquisition attacks, of which 5 failed — in every failing case a trusted entity could veto or refuse to certify. The successful Yuan Finance attack occurred despite the DAO having a vetoer who did not act.

  • Vetoes and delegation undercut the decentralization promise. 38 DAOs allow a referendum to be cancelled after proposal; 9 on-chain DAOs give veto power to the referendum initiator and 11 assign it to a veto authority. In the Venus case, a proposal with a majority of votes in favor of taking control of the treasury was cancelled by the vetoer. The authors note that while the vetoer mechanism improves security and delegation improves efficiency, both cripple decentralized governance.

  • Proposal rights are concentrated. Only Maker and Dxdao (#3 and #23) let anybody initiate a referendum; 19 DAOs use only voter-initiated referenda and 13 use only authority-initiated referenda, with every DAO using off-chain voting allowing authority-initiated proposals.

  • Voting platform split is even, with different trade-offs. 24 DAOs vote on-chain and 24 off-chain (all off-chain users rely on Snapshot). In Uniswap, which uses off-chain signaling before an official on-chain vote, off-chain votes draw larger voter turnout (participation is free), but on-chain votes see a much larger number of votes cast, since large holders are more likely to vote where it is binding.

  • Only one DAO shields votes. Every DAO except Illuvium shows votes as they are cast; Illuvium uses shielded (encrypted) voting off-chain, at the cost of being expensive to implement on-chain.

  • Certification is centralized off-chain, open on-chain. All DAOs with on-chain voting let anybody certify and execute a passed referendum; only Nexus Mutual, among on-chain DAOs, has a certification authority. Off-chain DAOs depend on a trusted multi-signature wallet to certify and execute, which can simply refuse to act. Cow Protocol and SuperRare instead use optimistic certification via oSnap (from UMA), where assertions are challenged in a court of a different protocol's token holders.

  • Execution delays are not universal. 9 DAOs with on-chain voting have no execution delay; the rest have at least 1 day and at most 3 days. 9 AAVE proposals have been cancelled after certification.

  • Referendum parameters vary widely. 11 DAOs have voting delays of 1 to 7 days (7 days for 0x), while 8 have none; Lido and Maker allow immediate voting and ENS, Radicle, and Cryptex start voting in the next block. All DAOs except Lido and Frax use "proactive" referenda requiring a quorum in favor; AAVE requires a vote differential of 80,000 votes, Curve requires a 51% threshold, and Maker is the only on-chain plurality/continuous-voting DAO. Angle, Origin, and Frax extend voting by 2 days if quorum is reached late, and Lido only allows votes against a proposal on the last day.

Methodology in Plain English

The authors picked 48 DAOs using explicit criteria: governance contracts on Ethereum, the largest treasuries as reported by DeepDAO in May 2024, "public" DAOs open to anyone (excluding private ones like the Graph Protocol), and at least one governance vote in the six months before May 2024 so that inactive projects were dropped. For each DAO they read the official documentation to find contract addresses and off-chain voting spaces, pulled the governance and governance-token contracts from Etherscan (which verifies deployed bytecode), ran an Ethereum archive node with the Erigon client to extract contract data, and collected off-chain voting data through the official Snapshot API. They then compared implementations dimension by dimension across the referendum lifecycle, and re-examined documented real-world attacks to see whether a different governance design would have blocked them.

Why This Matters

Impact on research: Prior work on DAO security focused on smart contract bugs, vote-buying, or the distribution of voting power. This paper redirects attention to the governance contract itself as an attack surface, showing that vulnerabilities persist even in bug-free code, and provides a reusable taxonomy of design dimensions plus a lifecycle framing that other researchers can build on.

Real-world applications:

  • DAO designers can use the lifecycle checklist (proposal rights, voting delay, vote-sniping defenses, execution delay, veto, certification) to audit their own governance contracts before deployment.
  • Token holders and delegates can identify which of the 7 flagged DAOs, or others with the same three flaws, currently expose their treasury to a Compound-style attack.
  • Multi-sig signatories and vetoers get a concrete lesson from the Yuan Finance and Compound cases: oversight only works if someone actually monitors and cancels malicious proposals.
  • Protocols choosing between on-chain and off-chain voting can weigh participation benefits against the risk that a trusted executor simply refuses to enact a legitimate result.

Industry relevance: With DeFi protocols securing more than 150 billion USD in total value locked (as cited in the paper), governance design failures translate directly into treasury loss and user risk. The finding that Compound responded to its attack by voting in a vetoer shows the trade-offs are already being negotiated in production governance.

Future Directions

  • Designing governance mechanisms that resist vote sniping and vote buying without reintroducing centralized vetoers or trusted executors — the paper notes current implementations can be "patched" to resist governance attacks, but the fundamental design tension remains.
  • Making private or shielded voting practical on-chain, since the only shielded-voting DAO in the study (Illuvium) does it off-chain and such schemes can be expensive on-chain.
  • Reducing dependence on external protocols for optimistic certification, which improves decentralization but hands dispute resolution to another protocol's token holders, with implications for autonomy.
  • Empirical follow-up on the 7 DAOs flagged as carrying all three flaws, and on whether adding oversight mechanisms actually deters attackers rather than merely reacting to them.

Target Audience

DAO designers and governance contract developers, DeFi security researchers and auditors, delegates and large token holders participating in governance, and policy or mechanism-design researchers studying decentralized decision-making. It is also accessible to informed crypto investors who want to understand the concrete risks behind governance votes in the protocols they use.

Authors’ abstract

A decentralized autonomous organization (DAO) is a governance entity that allows its stakeholders to manage blockchain-based protocols through smart contracts. The DAO explicitly specifies how stakeholders make and enforce decisions concerning a protocol's operation in a smart contract, aptly referred to as its governance contract. The design of this governance contract, therefore, has far-reaching implications for the security (trust) and privacy (transparency) of the smart…

The full abstract on arXiv
Read the original paper