On Exercising Governance Power in Decentralized Autonomous Organizations

The Hidden Flaws in Digital Democracy

How do thousands of anonymous stakeholders manage billions of dollars in digital assets without a central CEO? Decentralized Autonomous Organizations (DAOs) attempt to solve this. They use smart contracts—self-executing code on a blockchain—to automate decision-making. Researchers have studied these organizations to ensure their code is bug-free. However, a new study suggests that even perfect code can be exploited by the rules it follows.

The authors report that the structural design of these voting mechanisms can introduce a new class of vulnerabilities called "governance attacks." These are not software bugs in the traditional sense. Instead, they are exploits that leverage the fundamental logic of how a DAO proposes, votes on, and executes changes.

Designing the Rules of the Game

At its core, a DAO is a governance entity. Stakeholders manage blockchain-based protocols through encoded rules. To change how a protocol operates, a member must initiate a "referendum"—a formal proposal that is put to a vote. The goal is to achieve decentralized self-governance. In this ideal, no single person holds absolute power.

However, the authors argue that designing these systems involves difficult trade-offs. Designers must balance decentralization, security, and transparency. For example, if anyone can propose a change, the system is vulnerable to spam. If only a few "trusted" members can propose changes, the system becomes centralized. Much like a legislative body, a DAO must decide who can introduce a bill. It must also decide how the vote is tallied and who can veto a decision. The study finds that these choices create unique attack surfaces.

Mapping the Referendum Lifecycle

To understand where these vulnerabilities hide, the researchers categorized the governance of 48 major Ethereum-based DAOs. They used a structured lifecycle. As shown in, this process is divided into three temporal segments: the pre-voting period, the voting period, and the post-voting period.

During the pre-voting period, the study examines "admission controls." These are the rules governing who can start a referendum. The authors classify these into three types: "anybody" (anyone can propose), "voter-initiated" (requirers must hold a minimum amount of tokens), and "authority-initiated" (only specific, trusted addresses can propose). While authority-initiated models prevent spam, the authors note they erode the promise of true decentralization.

The voting period introduces further complexities regarding how votes are cast and tallied. The researchers distinguish between on-chain voting (recorded directly on the blockchain) and off-chain voting (using third-party platforms like Snapshot). The study finds a significant difference in participation. In Uniswap, for example, off-chain signaling votes see much higher voter turnout than official on-chain votes. This is likely because off-chain voting incurs no transaction costs .

Finally, the post-voting period covers "certification" and "execution." In on-chain systems, the smart contract typically handles these steps automatically. In off-chain systems, a "trusted" entity, such as a multi-signature wallet (a digital vault requiring multiple keys to open), must manually trigger the change. This creates a dependency on a third party. The authors describe this as a major trust trade-off.

Exploiting the Mechanism

The most striking result of the paper is the identification of "governance attacks." These succeed even when the underlying code is implemented perfectly. The authors use the July 2024 Compound DAO attack as a case study.

The attack relied on three specific design flaws: vote buying, voter apathy, and "vote sniping." In the Compound case, attackers used centralized exchanges and protocol loans to accumulate over 680,000 tokens .

Because the DAO lacked a "vetoer" (an authority capable of canceling a malicious proposal), the attackers pushed a proposal through.

Crucially, the attackers utilized "vote sniping." This involves waiting until the very last moments of the voting window to flood the system with votes. The authors report that in the Compound attack, 82% of the "yes" votes were cast within just 34 minutes of the voting period ending .

This sudden surge left other stakeholders with no time to react. This effectively hijacked the democratic process.

The study concludes that many DAOs remain vulnerable. The authors identify seven DAOs—including Uniswap, Gitcoin, and Radicle—that possess all three fatal flaws. These DAOs allow easy token acquisition, lack protections against vote sniping, and lack oversight mechanisms like vetoers.

Beyond Code Audits

This research shifts the focus of DAO security. It moves from mere software debugging to architectural design. It implies that a "bug-free" audit is insufficient. The governance logic itself can still be flawed. We can see that decentralization is not a binary state. It is a spectrum of trade-offs. Increasing security through vetoers or execution delays inherently erodes the autonomy of the organization.

For practitioners building these systems, the findings suggest defensive patterns. These include implementing execution delays. Such delays give users time to withdraw funds before a passed proposal takes effect. Developers should also consider creating "guard rails." These include veto authorities to intercept clearly malicious proposals.

Limits of the Analysis

While the study provides a comprehensive taxonomy, it is not exhaustive. The authors focus exclusively on public DAOs. Private DAOs, where decision-making is restricted to a known subset of stakeholders, were excluded. Therefore, the findings regarding the tensions between decentralization and security may not apply to closed governance models. Additionally, the study does not explore legal or regulatory implications. It focuses strictly on the technical and mechanical aspects of the smart contracts.