DAOs Push Protocols to Choose Code or Emergency Brakes

In July 2024 Compound’s governance process almost transferred 499,000 COMP, about $24 million at the time. Proposal 289 passed after a late surge of delegated votes: supporting addresses cast 563,591 votes in the final 34 minutes, and a final large block arrived eight minutes before the deadline. The on-chain mechanisms executed as written, but Compound had no emergency authority to pause the action. The allocation was later canceled and the protocol added a veto role that can halt execution.

DAOs set rules for registration, staking, delegation and voting that determine who can propose measures, who can vote and how outcomes are put into effect. Proposals commonly start in discussion forums or informal polls and then move to binding votes, either on-chain or through off-chain services. Once a vote meets quorum and approval thresholds, a smart contract, multisignature wallet or named executor carries out the result.

Two 2026 studies examined 48 large Ethereum DAOs and measured how those procedural rules concentrate voting power. Thirty-six of the 48 DAOs required some form of registration. Among those 36, the average share of outstanding tokens actually registered to vote was 21 percent. Centralized exchanges held on average more than 10 percent of outstanding tokens across the sample, and DeFi contracts held roughly 3.5 percent.

Fifteen DAOs required tokens to be locked before voting, with a median of 27.4 percent of supply locked. Some protocols offered stronger voting power for longer locks, up to four years. Services that lock tokens on behalf of customers and issue tradable substitutes concentrated voting power in a few entities: Convex accounted for roughly 53 percent of Curve’s voting power and about 46 percent of Frax, StakeDAO represented about 57 percent of Angle’s voting power, and Aura held about 65 percent of Balancer’s voting power in the measurements reported.

Delegation also concentrated authority. The ten largest holders controlled more than half of voting power in 39 of the 48 DAOs studied. Delegated voting was typically more concentrated than direct voting, allowing a small number of wallets, custodians or professional delegates to command large blocs of influence.

The second study defined a governance attack as an actor using authorized procedures to achieve a harmful result and catalogued 28 incidents. Researchers judged 16 of those incidents preventable by alternate mechanisms. Six incidents involved contract bugs; ten involved buying or borrowing tokens to influence votes. In the Compound case they traced 563,790 COMP through four centralized exchanges and another 118,089 COMP borrowed through Compound. The wallets that supported Proposal 289 held only 853 COMP before building their position over several months.

Options identified to reduce the risk of rapid, unexpected outcomes included extending voting windows when large blocs appear, requiring longer staking periods to make rapid accumulation costly, or giving a trusted council or named role the ability to pause execution. Compound’s change created a human check that can block actions but places pause authority in a small group of actors.