In April 2022, an attacker drained approximately $182 million from Beanstalk Farms, an algorithmic stablecoin protocol, netting roughly $80 million after repaying over $1 billion in flash loans. They did not break a single line of code.

What they broke was something far more interesting: The governance mechanism itself.

This is the pattern that defines the most damaging “hacks” in decentralized finance, and, more broadly, in decentralized protocols of all kinds. Most of these attacks are not technical exploits. They are governance exploits in disguise, where capital or coordination substitutes for authority. Whoever controls the most tokens controls the protocol. And in decentralized systems, controlling tokens can be as easy as borrowing a billion dollars for thirty seconds.

Decentralized governance is often presented as a more equitable alternative to traditional systems, yet recent history shows it is vulnerable to several distinct attack vectors. Below are five primary methods used to exploit these protocols:

Five Vectors

1. Flash Loan Leverage Manipulation

Beanstalk’s attacker borrowed approximately $1 billion in assets, primarily via flash loans from Aave, and used them to acquire STALK governance power through Beanstalk’s Silo, gaining approximately 79% of total voting power in a single transaction. They had pre-submitted a proposal disguised as a charitable donation to Ukraine. It contained a hidden function designed to execute immediately upon achieving a supermajority, bypassing the standard governance timelock. They voted. The system executed. They repaid the loan from the stolen assets. The entire attack took place within a single blockchain block, a matter of seconds.[1]

The design flaw was not in the code; it was in the governance design. An emergency mechanism with no delay is, by definition, a weapon for whoever can momentarily acquire a supermajority.

Worth noting explicitly: The Beanstalk attack combined two separate vectors simultaneously. The attacker used Flash Loan Leverage Manipulation to acquire majority voting power, then exploited Emergency Mechanism Abuse, specifically the protocol’s emergency proposal path, which carried no execution delay, to act on that power before any defensive response was possible. The same attack, the same transaction, two distinct governance vulnerabilities. (The Emergency Mechanism Abuse dimension is addressed further in that section below.)

Other Known Cases

  • This vector remains rare in execution precisely because the Beanstalk attack prompted widespread adoption of defensive countermeasures. Compound, Aave, and Uniswap all implemented voting blocklocks by 2021-2022, requiring tokens to be held for a minimum period before they confer governance power. The rarity of subsequent flash loan powered governance attacks is a partial success story for the DeFi security ecosystem.

2. Malicious Proposal Injection

The Tornado Cash governance attack of May 2023 is arguably the most technically sophisticated governance exploit on record. An attacker submitted what appeared to be a routine proposal to the Tornado Cash DAO. It passed. Upon passage, the attacker deployed a replacement contract to the same address, exploiting a blockchain determinism feature known as CREATE2, and substituted the approved bytecode with a malicious version that manufactured 1,200,000 votes, more TORN than had been delegated across all legitimate governance participants at the time. The attacker used this manufactured supermajority to drain the treasury and seize unilateral governance control.[2]

The proposal had been inspected before the vote. The substitution happened after the vote. There was no mechanism to detect it.

Other Known Cases

  • Audius DAO (July 2022): An attacker submitted a governance proposal containing malicious code that, upon execution, transferred approximately 18 million AUDIO tokens, worth roughly $6 million at the time, to a wallet under the attacker’s control. The Audius team paused the protocol within hours and the attacker’s gains were largely blocked. [3]
  • Build Finance DAO (February 2022): An attacker accumulated sufficient BUILD governance tokens over time to pass a hostile proposal granting unilateral minting and treasury access, then drained approximately $470,000 from the treasury. [4]

3. Whale Capture and Persistent Control

Not all governance attacks happen in a single block. Some take weeks or months and never trigger a security alert.

When Justin Sun purchased a majority stake in Steemit Inc. in 2020, he inherited roughly 20-30% of all circulating STEEM tokens accumulated through so-called “ninja mining” at the chain’s launch. When the network’s existing witnesses attempted to freeze his stake pending community review, Sun coordinated with major centralized exchanges, including Binance, Huobi, and Poloniex, to vote their custodied customer deposits on his behalf, without those customers’ knowledge or consent. The exchange-backed votes installed Sun-aligned witnesses and handed him effective governance control of a network worth more than $200 million.[5] The community ultimately forked the chain to create Hive.

The episode revealed something consequential: Custodied deposits at centralized exchanges represent a massive, largely invisible reserve of governance power that can be activated by any actor with the right relationships. It requires no technical exploit whatsoever, only capital and coordination.

Other Known Cases

  • Compound Finance (July 2024): A pseudonymous whale known as “Humpy” accumulated sufficient COMP tokens to pass Proposal 289, which would have redirected roughly $24 million into a yield product controlled by the proposing group. The proposal passed despite vocal opposition. A negotiated truce followed.[6]
  • Curve / Convex Finance (2022-present): The “Curve Wars” describe an ongoing dynamic in which Convex Finance accumulated a commanding position in vote-escrowed CRV (veCRV) governance power by offering users incentives to deposit their CRV through Convex. By early 2022, Convex had locked approximately 200 million CRV tokens, giving it by far the largest single veCRV position and effective influence over Curve’s liquidity emissions. The accumulation was transparent and within the designed incentive system, yet it produced the same structural outcome as a hostile whale capture: A single external entity exercising dominant governance influence over a protocol it did not build.[7]

4. Voter Apathy Exploitation

DeFi governance participation is consistently, stubbornly low, typically 2-10% of circulating supply in any given vote. This is not precisely a design flaw. Setting quorum thresholds high enough to prevent minority capture risks governance paralysis. Setting them low enables minority takeover. Most protocols choose functionality and pay the governance security cost.

In July 2024, a pseudonymous whale known as “Humpy” led a coalition that passed Compound Finance Proposal 289, directing approximately 499,000 COMP tokens, roughly $24 million, into a yield product controlled by the proposing group. The proposal passed not because the community supported it, but because the coalition timed their votes carefully and the broader governance community, dispersed and generally inactive, did not mobilize in time to stop it.[8] The episode drew immediate comparisons to a governance attack: a coordinated minority, exploiting low base participation, ratified a proposal the broader community would have rejected at full turnout.

A truce was eventually negotiated before the tokens transferred. But the vote had already passed. The governance mechanism had already rendered its verdict.

Other Known Cases

  • Uniswap governance (2023): Proposal 31, authorizing Uniswap v3’s deployment on BNB Chain, passed on-chain with significant controversy over which bridge provider would receive protocol fee revenue. Andreessen Horowitz cast approximately 15 million votes in opposition, citing bridge security concerns, and still lost, illustrating both the concentration of governance power in large single holders and the challenge of mobilizing broad-based opposition within standard voting windows.[9]
  • The recent (April 2026) spate of SPAM proposals aimed at the Stargaze community is another example of bad actors probing for voter apathy vulnerabilities by launching multiple bogus “airdrop” proposals in a short period, hoping a distracted electorate fails to act decisively.

5. Emergency Mechanism Abuse

Avraham Eisenberg’s attack on Mango Markets in October 2022 began with oracle manipulation, not governance. Using two accounts, he inflated the MNGO token price artificially, generating approximately $500 million in paper unrealized profits, then borrowed against that collateral to drain the protocol treasury of approximately $114 million.[10]

Then he turned to governance.

Eisenberg submitted an emergency governance proposal: return roughly $67 million and keep $47 million as a “bug bounty.” The community voted to accept. They voted to accept because the alternative was recovering nothing. Governance became the exit mechanism of extortion. The protocol’s own consensus process was weaponized against the people it was meant to protect.

Eisenberg was subsequently arrested and charged by the U.S. Department of Justice, the SEC, and the CFTC for fraud and market manipulation. He was convicted in federal court in April 2024. In May 2025, U.S. District Judge Arun Subramanian vacated all three counts. Federal prosecutors filed a notice of appeal in July 2025.[11]

Other Known Cases

  • Beanstalk Farms (April 2022): As described in the Flash Loan Leverage Manipulation section above, Beanstalk’s attacker used the protocol’s emergency proposal path, which carried no execution delay, as an essential component of the exploit. The emergency mechanism was not the entry point, but it was the weapon that made same-block execution possible before any defensive response could occur.

The Deeper Problem

These five vectors share a structural root: token-weighted governance is, mathematically, plutocracy.

Capital buys votes. More capital buys more votes. Unlike traditional corporate equity, which is accumulated over years and subject to regulatory disclosure requirements, governance tokens can be borrowed for seconds, purchased instantly, or accumulated covertly. The promise of decentralized governance frequently obscures extreme concentration of actual control.

Several mitigations exist for this structural problem, and many protocols have adopted them.

  • Vote-escrow models, pioneered by Curve Finance, require token holders to lock their governance tokens for fixed periods, up to four years, to receive proportional voting power, aligning governance influence with long-term commitment rather than momentary capital deployment.
  • Quadratic voting, which scales voting power with the square root of tokens held, reduces the marginal advantage of large holders; Gitcoin Grants has operated at scale on this model.
  • Delegation systems allow passive holders to assign their votes to active, knowledgeable participants, increasing effective participation without requiring every holder to engage directly.
  • Snapshot blocklocks, which require tokens to be held for a minimum period before they confer governance power, specifically address the flash loan attack surface. These are real improvements, adopted widely after the attacks described above. None of them fully resolves the underlying incentive structure: In a system where governance power is proportional to financial stake, sufficiently capitalized actors can still acquire controlling influence, just more slowly and at greater cost.

The situation creates a transparency paradox. All governance proposals, token balances, voting histories, and on-chain actions are publicly visible. Defenders and attackers have identical information. But attackers benefit from that visibility in ways defenders cannot: they study participation patterns, identify low-turnout windows, inspect proposal bytecode, and time attacks precisely. A rapid collective defensive response requires coordination infrastructure that most governance communities simply do not have.

Protocols have developed partial countermeasures for this as well. Timelocks, mandatory delays between a proposal’s passage and its execution, give communities time to detect and respond to malicious or unexpected outcomes before they take effect. Security councils, such as the multisig committees used by Arbitrum and other major protocols, can veto or pause proposals within the timelock window, providing a responsive human check on automated execution. Circuit breakers can pause protocol activity when anomalous transaction patterns are detected. Private voting mechanisms, using zero-knowledge proofs such as the MACI framework developed by the Ethereum Foundation, can prevent targeted voter suppression by concealing individual votes until after a defined period. These tools are meaningful. They address specific attack surfaces. What they do not address is the deeper problem of what happens when the attack is not anomalous, when it looks, by every on-chain metric, like a legitimate governance action.

The deeper philosophical challenge is what DeFi practitioners call “code as law.” Beanstalk’s attacker did nothing that was technically prohibited by the protocol’s own rules. The executed transaction was, by the protocol’s own logic, the legitimate outcome. There is often no legal or technical recourse after a governance exploit succeeds. Post-Eisenberg, regulators in the United States have shown willingness to pursue governance exploiters under fraud and market manipulation statutes, an important development, though legal frameworks remain nascent, jurisdictionally fragmented, and slow relative to on-chain attack execution. Timelocks, multisigs, and security councils are genuine improvements. They do not change the fundamental dynamic: The same principle that eliminates the need for trust in intermediaries becomes a shield for those who exploit the systems meant to replace them. Governance in decentralized protocols remains, at bottom, a political problem, and political problems do not have purely technical solutions.

A Closing Thought

The framing of “hacks” as technical failures is, at least to me, a kind of comfortable fiction. It lets us imagine that security is primarily an engineering problem: patch the code, audit the contract, add the timelock. But the attacks that have caused the most damage in decentralized governance are not engineering failures. They are failures of political design.

Building resilient decentralized organizations requires the same things that resilient governance has always required: Meaningful checks on concentrated power, information symmetry, genuine participation, and mechanisms designed with the assumption that not every actor will behave in good faith. The history of decentralized finance reads less like a story about the future of code, and more like a very old story about power.

Notes

  1. Beanstalk Farms, “Beanstalk Governance Exploit,” bean.money, April 19, 2022, https://bean.money/blog/beanstalk-governance-exploit.

  2. Halborn Security, “Explained: The Tornado Cash Hack (May 2023),” Halborn Blog, May 2023, https://www.halborn.com/blog/post/explained-the-tornado-cash-hack-may-2023.

  3. Bill Toulas, “Hackers steal $6 million from blockchain music platform Audius,” BleepingComputer, July 26, 2022, https://www.bleepingcomputer.com/news/security/hackers-steal-6-million-from-blockchain-music-platform-audius/.

  4. The Block, “Build Finance DAO suffers ‘hostile governance takeover,’ loses $470,000,” February 2022, https://www.theblock.co/post/134180/build-finance-dao-suffers-hostile-governance-takeover-loses-470000.

  5. CoinDesk, “Steem Community Plans Hostile Hard Fork to Flee Justin Sun’s Steemit,” March 17, 2020, https://www.coindesk.com/tech/2020/03/17/steem-community-plans-hostile-hard-fork-to-flee-justin-suns-steemit.

  6. The Block, “$24 million Compound Finance proposal passed by whale over DAO objections,” July 2024, https://www.theblock.co/post/307943/24-million-compound-finance-proposal-passed-by-whale-over-dao-objections.

  7. Nat Eliason, “Field Guide to the Curve Wars: DeFi’s Fight for Liquidity,” Every.to, February 3, 2022, https://every.to/almanack/curve-wars.

  8. CoinDesk, “COMP Token Rises as Whale Backs Down on Supposed ‘Governance Attack’ on Compound,” July 30, 2024, https://www.coindesk.com/business/2024/07/30/comp-token-rises-as-whale-backs-down-on-supposed-governance-attack-on-compound.

  9. CoinDesk, “Uniswap Vote on BNB Deployment Ends With Silicon Valley’s A16Z on Losing Side,” February 10, 2023, https://www.coindesk.com/tech/2023/02/10/uniswap-vote-on-bnb-deployment-ends-with-silicon-valleys-a16z-on-losing-side.

  10. CoinDesk, “How Market Manipulation Led to a $100M Exploit on Solana DeFi Exchange Mango,” October 12, 2022, https://www.coindesk.com/markets/2022/10/12/how-market-manipulation-led-to-a-100m-exploit-on-solana-defi-exchange-mango.

  11. CoinDesk, “Mango Markets Exploiter Avi Eisenberg Found Guilty of Fraud and Manipulation,” April 18, 2024, https://www.coindesk.com/policy/2024/04/18/mango-markets-exploiter-avi-eisenberg-found-guilty-of-fraud-and-manipulation.