The Governance You Can’t Flash-Loan: Bitcoin and Ethereum

In the second week of September 2026, an attacker pointed a freshly funded wallet at the treasury of the 1inch DAO. A week later, another drained roughly $9.4 million from the decentralized exchange Astroport and the staking protocol Drop by pushing a single proposal through the on-chain vote of their host network, the Cosmos app-chain Neutron. Neither attack broke a line of code. Both used the governance system exactly as it was designed to be used.

That is the through-line of every governance attack HOGE Wire has tracked this year. BonkDAO lost about $20 million in July after an attacker spent roughly $4.4 million to buy just enough votes. Term Finance lost $8.5 million in August despite having a timelock. Each raid needed the same thing first: a governance token liquid enough to accumulate. Buy the token, and you buy the votes. We covered the DAO side of that story in Silence Is a Yes.

Now look at the two networks that matter most. Bitcoin traded near $85,000 in late September and Ethereum near $2,700, together worth more than $2 trillion, eight-month highs and a far bigger prize than any DAO treasury on the planet. Neither has a governance token. Neither holds a vote you can win. There is nothing to flash-loan. That is not the same as being safe. It just moves the attack surface off the ballot and onto people, code and the chain itself.

What every 2026 governance attack had in common

In 2022, a trio of researchers at a16z crypto wrote the sentence every attacker has been living by since. A governance attack pays off, they argued, whenever the value an attacker can seize is greater than the cost of acquiring the votes plus the cost of executing the attack. Put the whole thing in one line: profit equals the prize, minus the price of a majority, minus gas. When a treasury is large and turnout is low, that math turns positive fast, and the researchers warned that markets cannot tell an honest buyer from an attacker, because both are simply buying tokens at rising prices.

Every 2026 incident is a data point on that equation. Neutron’s attacker bought a controlling block of the chain’s NTRN token, submitted a proposal, and pushed it through in minutes. BonkDAO’s spent a few million dollars to clear a quorum that only seven wallets bothered to contest. The security firm Blockaid counted at least seven governance takeovers between June and August, draining more than $22 million across Ethereum, Solana and Base, with zero smart-contract exploits among them. The code did what it was told. The voters told it to.

The reason this works is baked into what a governance token is. As Vitalik Buterin put it in his 2021 essay on coin voting, such a token is a bundle of two rights: an economic stake in a protocol and the right to steer it. Those two rights, he wrote, are “very easy to unbundle.” A flash loan borrows the voting power for a single block with no lasting economic interest at all. A cheap, thinly held token lets an attacker buy control for far less than the assets that control commands. The vote is the exploit because the token turned the vote into a commodity.

Bitcoin and Ethereum do not have a vote

Here is the part newcomers find genuinely strange: there is no button, anywhere, that lets BTC or ETH holders vote on the rules of Bitcoin or Ethereum. Owning coins buys you exposure to the price. It does not buy you a say in the protocol. The block size, the issuance schedule, the opcodes, the fee market, none of it is decided by a token-weighted poll. It is decided by what people choose to run.

Both networks govern themselves through a process borrowed from the early internet, called rough consensus. Anyone can write a proposal: a Bitcoin Improvement Proposal, or BIP, first formalized by Amir Taaki in 2011, or an Ethereum Improvement Proposal, an EIP. It gets debated on mailing lists, forums and GitHub. Client developers decide whether to write the code. Then the only vote that counts is the one taken with software: node operators, miners and validators either upgrade or they do not. Rough consensus is not unanimity, and it is not majority rule. It means an objection has to be answered on its merits, not simply out-voted.

The ultimate expression of that model is the fork. If a faction wants a different set of rules, it cannot outspend the others in a poll; it can only split off, run its own software, and let the market decide which chain is worth more. That is why the a16z profit equation, so lethal to DAOs, barely applies here. There is no line item for the cost of acquiring voting power when voting power is not for sale. An attacker who wants to bend Bitcoin or Ethereum has to find another door.

The four estates: where power actually sits

If no one votes, who governs? In practice, power over a tokenless chain is split across four groups, each holding a different lever and each vulnerable in a different way. Think of them as the four estates of a base layer. An attacker does not need all four; capturing any one of them, deeply enough, bends the protocol. Mapping them is the whole game, because the defense for each is different too.

Attack surface one: capture the maintainers

The first door is the smallest group with the most leverage: the people who decide what goes into the software. Bitcoin has no chief executive. Satoshi Nakamoto walked away in 2011, and a rotating handful of maintainers hold merge rights to Bitcoin Core, but no one of them can change the rules alone; a change has to survive a diffuse, argumentative, and famously conservative community. That diffusion is a feature. There is no single throat to grab. It is not, however, the same as having no point of leverage.

Ethereum is more concentrated, and one of its own builders said so on the record. Péter Szilágyi, an Ethereum Foundation core developer since 2015 who leads Geth, the network’s dominant execution client, wrote in a mid-2024 memo that later became public that “Ethereum may be decentralised, but Vitalik absolutely has complete indirect control over it.” Buterin’s “attention, direction of research, brainpower, donations and investments absolutely define which projects succeed,” he wrote, adding that winning Foundation backing was largely a matter of getting the right 5 to 10 people around Buterin onside. Buterin has long argued for pushing decisions away from any such inner circle, but the memo is a rare admission of how informal power actually clusters.

You do not need a token to exploit that. You capture the roadmap. Fund the research that suits your goals, cultivate the few contributors who set direction, get your preferred change onto the agenda while a rival’s quietly stalls for another year. It is slow, deniable, and it looks exactly like ordinary open-source politics, which is precisely why it works; the same indistinguishability a16z described for token buyers applies to influence. The defense is the thing that also makes the process maddening: many independent reviewers, more than one implementation of the protocol, and a culture that treats even a senior maintainer’s proposal as the opening of an argument rather than the end of one. Capture the maintainers and you can steer the conversation. You still cannot force anyone to run the code.

Attack surface two: poison the software supply chain

The sharpest tokenless attack is not political at all. It targets the code itself. Every node on Bitcoin or Ethereum runs a piece of client software, and whoever controls what ships inside that software controls what the protocol does. You do not need a majority of anything if you can slip a change into a binary that tens of thousands of machines will run without reading it line by line.

The template exists, and it very nearly worked in the wider open-source world. In March 2024, a Microsoft engineer named Andres Freund, chasing a fraction-of-a-second slowdown in SSH logins, stumbled onto a backdoor buried in XZ Utils, a compression library used across Linux. The attacker, operating a GitHub identity called “Jia Tan,” had spent roughly two years making legitimate-looking contributions, earned co-maintainer status from a burned-out volunteer, and, with help from sockpuppet accounts that pressured the original maintainer to step back, shipped code that would have let anyone holding a secret key run commands on a vast share of the world’s servers. It was catalogued as CVE-2024-3094 and rated the maximum severity score. It was caught by luck.

Bitcoin Core and Geth are exactly the kind of high-trust, thinly staffed open-source projects that scenario was written for. A backdoor in a widely run client would be a governance attack by other means: it changes what the network does without a single vote. The defenses are structural, not political. Ethereum deliberately runs several independent implementations, so no one codebase is the whole network; the client-diversity effort exists precisely because if a single client used by more than two-thirds of validators shipped a critical bug, it could stop the chain from finalizing. Bitcoin Core ships deterministic, reproducible builds, so independent parties can confirm the released binary matches the public source code, shrinking the space where a malicious build could hide. Neither measure is a guarantee. Both raise the cost of a supply-chain attack from compromising one person to compromising many, out in the open.

Attack surface three: split the chain

The third door is the loudest. If you cannot capture the maintainers or the code, you can try to split the community and walk off with the valuable half. A contentious hard fork is the closest a tokenless chain comes to a hostile vote, except the ballot is counted by the market afterward, in the price of each resulting coin.

Ethereum’s founding trauma is the case everyone cites. In 2016 an attacker drained about 3.6 million ETH, roughly a third of the funds held by an early experiment called The DAO. The community faced a choice: honor the principle that “code is law” and let the theft stand, or hard-fork the chain to claw the money back. It forked, at block 1,920,000 in July 2016, and the minority who refused kept running the old chain as Ethereum Classic. The lesson has echoed ever since: the social layer can override the code, which is reassuring when the code has just been robbed and unsettling when you imagine who else might one day invoke it.

Bitcoin’s defining fight was about power, not theft. The Blocksize War of 2015 to 2017 pitted businesses and miners who wanted larger blocks against users and developers who did not. In 2017 more than fifty companies signed the New York Agreement to force the change through; users answered with a user-activated soft fork, BIP148, whose August 1 flag day some still call Bitcoin Independence Day. SegWit activated, the corporate hard fork was abandoned that November, and the faction that wanted big blocks split off as Bitcoin Cash. The conclusion Bitcoiners drew, and that still governs the culture, is blunt: the economic full nodes are sovereign, not the miners and not the exchanges.

The live version is quieter and unresolved. Bitcoin Core’s version 30, released in October 2025, dropped a long-standing default limit on how much arbitrary data a transaction can stuff into an OP_RETURN output. Bitcoin Knots, a rival client maintained by Luke Dashjr, keeps a strict filter and treats the extra data as spam. Crucially, this is a disagreement about what nodes choose to relay, not about which blocks are valid, so, as Bitcoin Magazine lays out, the chain does not split; both clients accept the same blocks. Even so, Knots’ share of reachable nodes climbed from around 18 percent in 2025 and kept rising through 2026 as the argument pulled in more operators. There is even a bid to harden the policy fight into a consensus rule, a proposed soft fork floated as BIP-110 that would temporarily cap that data at the protocol level, though miner signaling for it has sat near zero. The fork threat stays loaded even when no one pulls the trigger.

Attack surface four: collude at the miner or validator layer

The fourth door is the one security researchers have modeled for over a decade: the 51 percent attack, where whoever controls a majority of block production can censor transactions, reorder them, or reverse recent ones. On a tokenless chain you cannot win this with a governance vote, but you can buy it with hardware or with stake, and in practice you do not even need a clean majority in one hand. You need a cartel that can coordinate.

Bitcoin’s block production is startlingly concentrated for a system sold on decentralization. Individual miners point their machines at pools, and by late September 2026 the two largest, Foundry USA and AntPool, dominate the field, and the top four pools together control more than 70 percent of it. The saving grace is that pools are not the miners themselves; a pool operator who abused its position would watch its hashrate walk to a competitor within hours, and full nodes would still reject any invalid blocks it produced. The structural fix is Stratum V2, a protocol that hands transaction selection back to individual miners instead of the pool, and by May 2026 pools representing roughly 75 percent of hashrate had signed on to an open standard for it. We have written about how misleading the pool leaderboard can be, in The Leaderboard Blind Spot, and about who actually secures the network.

On Ethereum the equivalent worry is staking concentration. Lido, the largest liquid-staking protocol, controls roughly a quarter of all staked ETH, down from a peak near a third in 2023 but still enough that a single set of node operators sits close to the thresholds where it could, in theory, censor or stall the chain; The Block reported the protocol consolidating some $16 billion of staked ETH in mid-2026. Lido’s own answer is a dual-governance design that lets holders of its staked-ETH token veto decisions made by holders of its LDO governance token, an unusual admission that controlling that much stake is itself a governance risk. Either way, the attack here is not a proposal anyone submits. It is accumulation, quiet enough that no vote ever records it.

The crossover case: how Justin Sun almost bought a blockchain

The cleanest illustration of these surfaces colliding happened on a smaller chain in 2020, and it is worth studying because it shows both the attack and the only defense that reliably beats it. In early 2020 Justin Sun’s Tron acquired Steemit, the company behind the Steem social blockchain, along with a large pre-mined stake of Steem tokens. The community feared he would use that stake to control the chain’s elected validators, known as witnesses, and moved to freeze the pre-mined coins.

Sun’s countermove is the part every governance designer should memorize. He persuaded three major exchanges, Binance, Huobi and Poloniex, to stake their customers’ deposited Steem, tens of millions of tokens that belonged to users rather than to the exchanges, and to vote out the community’s witnesses and install his own. It was a governance takeover executed with other people’s coins, and for a few days it worked.

Then the defense that has no equivalent inside a DAO kicked in. After a public backlash the exchanges pulled their votes, and the community exercised the ultimate exit: it hard-forked the chain into a new network, Hive, that simply excluded Sun’s stake from the opening balances. The bought majority was left holding a chain its users had walked away from. That is the through-line of tokenless governance. You can capture the validators, but you cannot stop the community from forking away and taking the value with it. The process is messy, contentious and slow, and it is the reason coin-buying alone rarely captures a base layer.

Why you cannot just buy Bitcoin’s governance

Step back and the pattern is clear. The DAO attacks work because a token fuses two things that ought to be separate: the right to profit from a protocol and the right to control it. Buterin’s argument in 2021 was that this fusion is fragile, because the two rights are trivial to pull apart. A lender rents you the governance power for a single block; an attacker buys a controlling stake in a token worth far less than the treasury it commands. Either way, the person steering the protocol has little or no stake in its long-term value.

Bitcoin and Ethereum refuse that fusion at the base layer. Holding the coin is pure economic exposure; steering the protocol runs through code, nodes and rough consensus. There is no token whose purchase confers control, so there is nothing to unbundle and nothing to flash-loan. When the industry’s own investors survey the wreckage of a decade of coin voting, this is the conclusion they reach. As Ali Yahya, a general partner at a16z crypto, put it, “We spent the last 10 years rediscovering the hard way that direct democracy is a bad idea.” The base layers never tried direct democracy in the first place.

The cost of that refusal is legibility. A DAO vote is ugly but readable: you can see the quorum, the tally, the exact wallet that tipped the balance. Rough consensus is the opposite. There is no scoreboard, which is precisely why informal power, the maintainers, the foundation, the pools, the funders, matters so much, and why Szilágyi’s memo landed as hard as it did. Bitcoin and Ethereum traded a clean, buyable attack surface for a murky, human one. That is an excellent trade against a flash-loan raider. It is a far worse trade against a patient, well-funded influence campaign that never has to file a single proposal.

The live tests of 2026: OP_RETURN, timewarp and the roadmap

Governance you cannot buy still has to actually work, and 2026 is stress-testing both networks in public. On Bitcoin, the OP_RETURN dispute is one live wire and the difficulty adjustment is another. Bitcoin carries a long-known consensus bug called the timewarp flaw, which a colluding miner majority could abuse to distort block timing, and a proposed cleanup soft fork would finally close it. The argument over whether and how to ship that fix is the BIP process working on a genuine vulnerability rather than a values clash, and we walked through the whole fight in Bitcoin’s one real bug.

Ethereum’s process runs on a faster clock. Its Fusaka upgrade activated in December 2025, and the next hard fork, Glamsterdam, is targeted for later in the roadmap with changes to how blocks get proposed. None of it is decided by a vote of ETH holders. It is hashed out on All Core Devs calls, in EIP threads and in client code, then shipped only when the client teams and node operators agree to run it. Even a change as consequential as a new transaction type, such as the account-abstraction feature EIP-7702 that shipped in 2025, moves through that same rough-consensus pipeline; we unpacked its security trade-offs in the delegation you forgot to revoke.

The attack, in both cases, is to short-circuit that pipeline: to ram a change through without genuine consensus, or to stall a needed one until the network ossifies around a known flaw. That is where the maintainer-capture surface and the roadmap meet. It is also why the culture prizes friction. A slow, argumentative, hard-to-move process is exactly what a governance attacker hates, because there is no quorum to buy and no timelock to wait out, only a crowd that has to be genuinely convinced.

Where tokenless governance actually fails

None of this makes Bitcoin and Ethereum well-governed in any comfortable sense. It makes them hard to capture with money in a single transaction, which is a narrow and valuable property, not a general one. The failure modes are real, and they are the mirror image of the DAO failures. The problem is not that these networks are too legible and too buyable. It is that they are too illegible and too informal.

The clearest gap is developer funding. The people who maintain Bitcoin Core and Ethereum’s clients are paid by someone: exchanges, foundations, corporate labs, grant programs. Whoever writes those checks has a soft hand on the roadmap, and the arrangements are far less transparent than an on-chain treasury. Capture the payroll and you get much of what maintainer-capture promises, legally and without a headline. This is the surface that no timelock and no fork can reach, because it operates entirely off-chain.

The second gap is apathy at the node level, which is what turns the supply-chain surface from theory into risk. Most users run defaults and upgrade when they are told to. That means whoever ships the default client, and whatever that client chooses to relay or filter, wins by sheer inertia; the Core-versus-Knots standoff is partly a fight over exactly that default. A newer form of quiet weight is arriving on top of it: spot ETFs and large custodians now accumulate BTC and ETH on behalf of clients, which means that in a future contentious fork a small number of regulated intermediaries could hold the swing weight, effectively deciding which chain keeps the ticker without ever asking the people who actually own the coins.

Finally, there is the regulator-shaped hole. Because a base-layer protocol has no company, no board and no legal personality, the tools that reach a DAO or a token issuer do not reach it. The SEC and its overseas counterparts regulate the on-ramps, the exchanges, custodians and issuers, not rough consensus itself. That is a shield against a lawsuit aimed at the protocol, and it is a gap if a captured protocol ever does something a court would want to undo. There is simply no defendant to serve.

The defender’s read: two very different attack surfaces

The most useful thing a builder or investor can do is figure out which of two worlds a given protocol lives in, because a defense from one is close to useless in the other. A timelock is a fine brake on a token vote and completely irrelevant to a poisoned build. Client diversity protects a base layer and does nothing for a DAO treasury governed by one liquid token. The two attack surfaces barely overlap.

Most of crypto is a hybrid of the two columns: an application with a token vote, running on a base layer with rough consensus, held through intermediaries with their own concentration. Each layer has its own door, and each door needs its own lock. The mistake is assuming that because the base layer cannot be flash-loaned, the whole stack is safe.

What it means for holders and builders

For a holder, the reassuring headline is that the value of your BTC or ETH is not one lax quorum away from being voted into an attacker’s wallet, the way a DAO treasury can be. There is no governance token to buy, no proposal that can seize the coins, and no timelock to race. The base layer’s attack surface is slow and social, which buys time and usually leaves a fork as an escape hatch of last resort.

For a builder, the lesson is to know which layer you are defending and to stop borrowing the wrong playbook. If your protocol puts real value behind a liquid token vote, you live in the DAO column, and timelocks, high quorums, past-block snapshots and live monitoring are your friends. If you are building or running base-layer infrastructure, your risks are client concentration, unaudited dependencies, opaque funding and the slow capture of a roadmap, and not one of those is solved by a governance dashboard.

A short set of questions maps almost any protocol’s governance attack surface:

• Does real value sit behind a token vote that anyone can buy into, and if so, how cheap is a majority relative to what it controls?
• How concentrated are the validators, miners or client implementations, and how quickly could a few parties coordinate?
• Who holds commit or merge authority over the software, how many of them are there, and who pays them?
• Are the builds reproducible, so the shipped binary can be checked against the public source?
• If governance were captured tomorrow, is there a credible fork or exit, and who would keep the ticker and the price?

The through-line of a brutal year of governance attacks is that the vote was never the safe part. It was the target. Bitcoin and Ethereum sidestepped that trap by refusing to hold a vote at all, and they paid for it with a messier, more human, less legible way of deciding things. You cannot flash-loan it. You can still, given enough time and money and patience, try to capture it. The defense, in the end, is the same one the Steem community reached for and the same one the Blocksize War settled: a large, alert and stubborn set of people who will fork away from anyone who tries.

Frequently Asked Questions

Can you flash-loan attack Bitcoin or Ethereum governance?

No. A flash-loan governance attack works by borrowing a protocol’s governance token to win an on-chain vote, then repaying the loan in the same transaction. Bitcoin and Ethereum have no governance token and hold no such vote; their rules change through rough consensus among developers, node operators and miners or validators. There is nothing to borrow and no ballot to win, so the flash-loan playbook simply does not apply to the base layer.

What is a governance attack?

A governance attack is when someone uses a protocol’s own decision-making system, rather than a software bug, to make it act against its community’s interest, usually to drain a treasury or seize control. In 2026 the common form was buying or borrowing enough of a DAO’s governance token to pass a malicious proposal, as happened to BonkDAO, Term Finance and Neutron. The code behaves correctly; the attacker simply controls the votes.

Who actually controls Bitcoin?

No single person or company controls Bitcoin. Satoshi Nakamoto left in 2011, and changes to the rules require rough consensus among a diffuse group: the developers who maintain the software, the miners who produce blocks, and above all the economic full nodes that decide which version of the rules to enforce. The Blocksize War of 2015 to 2017 established that these node operators, not miners or big companies, hold the final say.

Is a 51 percent attack a governance attack?

It is a governance attack at the security layer rather than the political one. Instead of winning a vote, an attacker acquires a majority of mining hashrate or staked coins and uses it to censor or reverse transactions. It does not require a token ballot, which is why concentration matters: Bitcoin’s top four mining pools already control more than 70 percent of hashrate, and on Ethereum a single staking protocol controls roughly a quarter of all staked ETH.

Could a backdoor in Bitcoin Core or Geth take over the network?

In principle, yes, which is why it is treated as a serious governance risk. Because every node runs client software, malicious code merged into a widely used client could change what the network does without any vote, much as the 2024 XZ Utils backdoor nearly compromised large parts of the internet. The defenses are client diversity, running several independent implementations, and reproducible builds that let anyone verify the released software matches its public source code.

Anneke de Vries covers security and exploits for HOGE Wire.