Forty-three of forty-seven validators voted no, and 70 billion tokens were minted anyway

There is a question worth asking about anything you hold, and almost nobody asks it until the answer has already been used: if somebody wanted to change how many of these exist, who would have to agree?

It is not the same question as who owns the token, and the gap between the two is the subject of this article. On a chain where votes are weighted by stake, a holder with one coin and a holder with a billion are both holders and only one of them is a decision. On a token with a mint function, the entire electorate is whoever can satisfy the modifier on that function. Four cases follow, and they are ordered by how legitimate the process was — because the first one broke no rules at all, and that is the point of putting it first.

1. The vote that followed its own rules

In 2021 Cronos burned 70 billion CRO, cutting the supply from 100 billion to 30 billion — 70% of everything that existed, and at the time one of the largest burns anybody had done. In March 2025 Cronos Labs published a proposal to mint those 70 billion back, into a strategic reserve with a five-year vesting schedule, in support of a roadmap that included an ETF filing. Voting ran from 3 to 17 March.

Cronos@CronosApp

Today, an ambitious proposal was published for voting on the Cronos POS governance forum. The community is invited to vote on a bold plan to restore Cronos' Golden Age, with a massive investment allocated to support Cronos Roadmap (incl. its ETF) and America’s ambition to become https://t.co/7i06DnTx7T

While the vote was open, a holder counting validators published the tally. His figures: 43 of the 47 validators had voted no, and yes was nevertheless ahead — 45.82% of voting power, 1.26 billion CRO, against 44.43% and 1.22 billion. Four validators operated by Crypto.com had voted in favour, and they carried more stake between them than the other forty-three carried together. Quorum, at that point, was still short: 26% against a 33.4% requirement.¹

Anand Chennai2London@chennai2london

$CRO unburn of the 70 Billion token supply - voting ends tonight with a few hours remaining. I explained it in my post 2 weeks back. As it stands today nearly all of the validators and holders have voted No (43 out of 47). However the YES votes are leading with 45.82% (1.26 https://t.co/5oSy0GMvyP

His own summary of it: “nearly all of the validators and holders have voted No (43 out of 47)”, while “the YES votes are leading with 45.82% (1.26 billion CRO) to NO Votes with 44.43% (1.22 billion CRO)”, because “the 4 validators owned by @cryptocom have voted in favour”. The comparison he reaches for is an election in which 95% vote one way and the outcome goes the other because the remaining 5% hold more of what is being counted.

Reporting from Unchained reported the structural fact twice that month, and the two figures are worth keeping apart. Its piece during the vote put the holding at 80% and drew the inference from that number: Crypto.com could “theoretically pass this proposal entirely on its own”. Its piece after the vote widened it to “between 70-80% of the total voting power according to three separate sources”. The inference survives the lower bound — a simple majority of non-abstaining stake is enough under these rules — but it was drawn from the point estimate, not the range, and the flat 80% that circulates afterwards is the top of somebody else’s interval.

The near-tie did not last, and the timestamps say how narrowly it was caught. That post was published at 13:28 UTC on Sunday 16 March. The same outlet recorded what happened thirty-two minutes later: “At 14:00 UTC on Sunday when 3.35 billion CRO tokens were added to the pro category, achieving quorum and pushing the vote well over the top.” The final tally, read from the chain rather than from the coverage: 4,612,887,499.80 CRO in favour, 1,306,267,987.43 against, 1,491,565,221.63 abstaining and 8,128,853.01 vetoing — 62.18%, 17.61%, 20.11% and 0.11% of the 7,418,849,561.87 cast, on a turnout of 70.18% against a 33.4% quorum. The 61.18% that circulates in the reporting is a slip, and it is catchable without leaving the page it appears on: those four shares would sum to 99.01, which no set of shares does. A vote that had been within one and a half points at the midpoint finished more than forty points apart, and the distance was covered in a single afternoon by one block of stake. Whatever else is true about it, no rule was broken — the process ran exactly as specified, and the specification is what produced the result.

CoinBaron@CoinBaron

@zachxbt @kris Scam? No. Centralized? Yes. The governance structure of Cronos works with validators, where many CDC users delegate their funds to validators controlled by the company. CDC therefore has the majority of the voting power on behalf of their customers. Unfortunately CRO never has

The wording of that reply is worth keeping because of what it refuses to say: “Scam? No. Centralized? Yes.” The structure it then describes — customers delegating to validators the company runs, so the company “has the majority of the voting power on behalf of their customers” — is a description of a mechanism, not an allegation about anyone’s conduct, and this article uses it the same way.

One thing more, from the same request. The ballot is proposal 29 on the Cronos hub, and its type is a software upgrade: “Cronos New Golden Age to Onboard Billions and Invest in the US — Cronos POS V5 upgrade”. Its description runs to 621 characters, and the words that would say what it does are not among them. No “70 billion”. No “mint”, no “reissue”, no “supply”. No mention of the 2021 burn. What it offers instead is “updates to the network parameters and token allocations”, and a pointer to a discussion on GitHub.

That is checkable in one request and it sharpens the case rather than softening it. Every rule was followed; the ballot simply did not state its own subject, and the substance sat somewhere a vote cannot reach. Forty-three of forty-seven validators opposed a text that does not contain the number it changed.²

2. The vote that was run properly, and still came down to two holders

It would be convenient if the first case were simply a story about one company. It is not, and the cleanest demonstration is a vote almost nobody criticised. On 28 August 2026 Solana’s validators concluded SGP-0002, which doubled the rate at which the chain’s emission declines. It was the network’s first formal governance vote, the proposal was public for weeks, and it passed with 67.0% of participating stake against a two-thirds bar — clearing it by about a third of a percentage point.

Turnout was 60.7% of eligible stake, and that second number is the one that gets left out: because SGP-0002 was carried by 67.0% of a 60.7% turnout, the share of all eligible stake that actively voted for the change was 67.0% × 60.7%, or about 40.7% — roughly two fifths, not two thirds. Nothing improper happened; a threshold has to be measured against somebody, and measuring it against the stake that turned up is the normal choice. But “two thirds agreed” and “two fifths of the stake decided” describe the same vote, and only one of them is usually quoted.

One thing that product does not license, and it matters because the mistake is easy and flattering. The remaining three fifths did not oppose anything. There are three distinct positions in a vote like this and they have three different arithmetic consequences: a no counts against, an abstention enlarges the denominator without helping either side, and a validator that never voted is absent from both. Only the first is opposition. A turnout figure measures attention, not dissent.

Within that margin, individual holders were decisive. Kraken voted against through the count and switched at the last minute; its 8.92 million SOL is about 3.4% of the roughly 263 million that voted, against a winning margin of a third of a percentage point — on the order of 870,000 SOL. One holder was carrying ten times the distance between passing and failing. Galaxy moved from abstaining to in favour in the final hour. The vote was legitimate, well-run, closely contested and decided by a handful of participants, and all four of those things are true at once. That is the ordinary condition of stake-weighted governance rather than a failure of it. The emissions article covers what SGP-0002 actually changed.

3. The votes that were manufactured

On 20 May 2023, at 07:25 UTC, an attacker took control of the Tornado Cash DAO by submitting a proposal that appeared identical to one the community had already reviewed and approved. After it was voted through, the deployed code was replaced — the original contract was destroyed and a new one put at the same address — and the replacement granted the attacker’s addresses 10,000 votes each, roughly 1.2 million in total. Against the roughly 700,000 legitimate votes in existence, that was a majority, and a majority was all it needed to be. The attacker had not overwhelmed the electorate by some enormous multiple; they had simply become most of it, in one transaction, at no cost.

What that permitted is the part worth reading closely, because it generalises to any protocol built the same way. The security researcher samczsun enumerated it at the time: withdraw every locked vote, drain the tokens held by the governance contract, and disable the router. Then came the correction, and the correction is the more instructive half.

samczsun@samczsun

Correction: @CellierLael correctly pointed out that Tornado Cash Nova, deployed to Gnosis Chain, is a proxy that is administered by governance. Therefore, the attacker is also able to drain all of the ETH in that pool by upgrading the contract https://t.co/wO96ngrlm9

The correction reads: one of the pools “is a proxy that is administered by governance. Therefore, the attacker is also able to drain all of the ETH in that pool by upgrading the contract.” The first list was what the rules allowed; this was what the rules could be rewritten into.

One of the deployments was a proxy administered by governance. A proxy separates the address people interact with from the code that runs behind it, and whoever administers the proxy can replace that code. So governance control was not merely control of the parameters somebody had thought to make adjustable — it was control of the contract’s behaviour in general, including behaviour nobody had written yet. For a supply question this is the sharpest version of the problem: if the mint rule lives behind a proxy, the mint rule is whatever the proxy admin decides it should be tomorrow.

The ending is worth recording too, and it is not the one the numbers suggest. On 26 May control returned to the original holders on a vote of 517,000 in favour and none against — an outcome that reads like a governance system defeating an attack, and was not. The proposal was submitted by the attacker. Governance did not win anything back; it was handed back, through the same mechanism that had taken it, and it executed both instructions with equal obedience. The system worked perfectly on the way in and on the way out. It simply had no way to tell whose votes those were.

4. The votes that were rented

The fourth case removes ownership from the picture entirely. On 17 April 2022 an attacker borrowed more than a billion dollars through a flash loan — a loan that must be repaid inside the same transaction that takes it out — converted it into the assets Beanstalk’s governance accepted as stake, and deposited them. That produced roughly 79% of the total staked weight, on the post-mortem reconstructions by Immunefi and Veridise — a figure worth attributing, because several accounts print 67% here instead and that is the threshold being cleared rather than the share that was captured. An emergency-commit path in the governance module allowed a proposal to execute immediately on a two-thirds supermajority, so the malicious proposal was passed and executed and the loan was repaid, all inside one transaction. The protocol lost $182 million.

Nobody bought a governance token. The voting power was rented for a few seconds and handed back, which means the defence everybody assumes exists — that an attacker would have to acquire a majority stake and would suffer from destroying its value — did not apply. A supermajority requirement is a defence only when acquiring the supermajority is expensive and slow. Where stake can be assembled and returned atomically, the threshold is a price, and the price was one transaction fee.

The tokens with no vote at all

All four cases involved governance, which makes them dramatic and slightly unrepresentative. Most tokens have no vote of any kind, and their supply is controlled through something simpler.

1. 1

   A single key. A mint function guarded by an owner check, where the owner is one ordinary address. One signature changes the supply, with no notice, no proposal and no delay. This is the most common arrangement in existence.

2. 2

   A multisig. The same function behind an m-of-n wallet. The count that matters is m — not n, and not the number of names on the website. A 2-of-3 is two people, whatever the three are called.

3. 3

   A timelock. The same power, but the transaction must be announced and then wait. A timelock does not reduce who can act; it changes how much warning you get, which is a real and different benefit.

4. 4

   An upgradeable proxy. Whoever can upgrade can replace the mint rule, the cap, the pause switch and the ownership check together. Reading the current implementation tells you what the token does today, and the proxy admin tells you who decides what it does tomorrow.

5. 5

   Nothing at all. No mint path, no upgrade path, no admin. Rare, checkable in a couple of minutes, and the only configuration where the supply rule you read is the supply rule you get.

AccessSmart contract access utilities and implementationsOpenZeppelin Docs

Counting it, for any token

The useful output is a single number: the smallest set of independent parties whose agreement is sufficient. It is worth writing down before there is a reason to care, because every case above became interesting only after the power had been used.

1. 1

   Find every path that changes supply. A mint function, a burn-from function, a rebase multiplier, a chain-level issuance parameter, an upgrade entry point. Miss the upgrade path and the rest of the count is decorative.

2. 2

   For each path, read the guard rather than the documentation. A modifier, a role, a governance address. The guard is what executes; the documentation is what somebody intended when they wrote it.

3. 3

   Resolve each guard to people. A multisig resolves to its threshold, a governance contract to the stake needed to pass a proposal, an owner address to one key. Follow it until you reach something that cannot be reduced further.

4. 4

   For stake-weighted governance, get both numbers. The threshold and the participation. Multiply them, and compare the result with the largest holders — if any single holder exceeds it, the count is one.

5. 5

   Ask what warning exists. A timelock, a voting period, a public forum. Warning is not control, but it is the difference between reading about a change and being able to act on it.

What the ledger shows afterwards

One figure of our own, for the case this article opens with. Cronos is in the index, and its twelve-month net float change is measured on the same daily pipeline as every other token.

Net float is emission plus unlocks minus burns, so this figure is not the re-mint on its own — the reserve vests over five years, and other things moved in the same window. What the number does establish is direction and order of magnitude: the float behind this token grew substantially over twelve months, which is the measurable consequence of a decision that a majority of validators, counted as validators, opposed. It is computed from the measured float series, and the individual flows listed on the token’s own page are a separate and thinner ledger — evidence of particular events rather than a decomposition of the percentage. Our net supply figures are withheld rather than estimated wherever a reading is missing.

The five mechanisms that add supply in the first place — and which of them a vote can reach — are set out in where new tokens come from. Whether destruction is permanent at all — the question the Cronos case raises and does not settle — is worked through in can burned tokens come back, and the six places a burn can send tokens in where burned tokens go. How every figure here is measured, including what we do when a reading is missing, is on the methodology page. Scheduled releases are on the unlocks page.