Proof of Reserves in 2026: Who Proves What, and What It Misses

The week proof of reserves stopped being a slogan

On 29 October 2026, attackers drained roughly $6.3 billion from Binance, the largest theft in the history of crypto and bigger than the $1.5 billion Bybit loss of early 2025. CoinDesk reported that the breach started with a fake job interview, spread through malicious developer software, and ended with an EIP-7702 delegation that turned a hot wallet into a sweeper the attackers controlled. Investigators at TRM Labs and Chainalysis tied it to the same North Korean group behind the Bybit raid.Within days a second headline chased the first. On 2 November, Binance published a fresh proof of reserves covering ten assets, each backed at least one to one, and pledged to absorb the loss itself rather than pass it to users. The message was blunt: the money left, but your deposits are still here, and here is the math. For a lot of readers that revived an older question that never really went away after FTX. What does a proof of reserves actually prove, and how far should you trust one?This is a comparison piece, not a ranking. The goal is to show how the main methods work, where they agree, and where each one quietly stops short. Prices here are in US dollars and the regulatory lens is American, because the United States is the one major market that still does not require any exchange to publish a proof of reserves at all. Bitcoin traded near $78,000 and Ether near $2,300 as the Binance story unfolded, according to Fortune, a reminder that the sums held in custody are enormous even on a quiet week.

Three different proofs hide inside one phrase

The phrase proof of reserves is doing the work of three separate claims, and most arguments about it come from mixing them up. The first is proof of assets: a demonstration that the exchange controls a certain pile of coins, usually by signing a message from the wallets that hold them or by pointing to public addresses. The second is proof of liabilities: a demonstration of how much the exchange owes its customers in total. The third, and the only one that means anything on its own, is proof of solvency: evidence that the assets are at least as large as the liabilities.Vitalik Buterin put the relationship plainly in an essay written in the days after FTX failed. “If you prove that customers’ deposits equal X (proof of liabilities), and prove ownership of the private keys of X coins (proof of assets), then you have a proof of solvency,” he wrote. The trap is that proof of assets by itself proves nothing useful. An exchange can show it holds a billion dollars in Bitcoin while owing its customers three billion, and the wallet signature will look just as convincing either way.So a credible proof of reserves has to join the two halves. Proving the coins exist is the easy part, and the part exchanges love to show off with a splashy dashboard. Proving the coins are not already promised several times over is the hard part, and the part that collapsed in every major failure of the last cycle. Keep that split in mind, because almost every method below is strong on one side and weaker on the other.

The Merkle tree: proving your balance is in the set

The workhorse of the liabilities side is the Merkle tree, a structure the analyst Nic Carter helped push into the mainstream after Mt. Gox and again after FTX. The exchange takes every customer balance, hashes each one into a leaf, then combines the leaves in pairs, hashing upward until a single value remains at the top: the Merkle root. Change any balance anywhere in the tree and the root changes. The exchange publishes the root and hands each customer the short path of hashes needed to prove their own balance was folded into it. You confirm that your number is in the set without ever seeing anyone else’s.This is the method Changpeng Zhao called for in November 2022, hours after Binance first moved on the collapsing FTX. Zhao proposed that every exchange publish a Merkle-tree proof of reserves and argued that crypto firms, unlike banks, should not run on fractional reserves. Kraken, OKX, Bitget, Bybit and others followed within weeks, and the Merkle tree became the industry default it still is in 2026. Nic Carter went on to write a widely cited primer, Proof of Reserves for Policymakers, that laid out both the promise and the holes.The method is genuinely useful and genuinely limited. It proves that the liabilities the exchange chose to include add up to a published total, and it lets you confirm you are counted. It does not prove the exchange included every liability. A platform can leave accounts out of the tree, or carry internal negative balances that net against customer deposits, and the root will still validate for everyone who checks. The tree is only as honest as the list fed into it, which is why the total has to be checked by someone outside the company. The same accountability gap haunts on-chain organizations that call themselves trustless but still lean on a handful of insiders to assemble the numbers, a tension we traced in our look at who actually answers for a DAO.

The zero-knowledge upgrade: solvency without the leak

The plain Merkle tree has a privacy problem and a soundness problem. On privacy, publishing the sum of all balances tells competitors exactly how big an exchange really is, and a careless design can leak individual positions. On soundness, a naive tree lets an exchange slip in fake negative accounts to shrink its apparent liabilities. The fix that took over by 2026 is the zero-knowledge proof. Binance moved first, adding a zk-SNARK layer in early 2023 built with the Polyhedra Network, and it still runs that stack today.A zk-SNARK lets the exchange prove a statement about the whole tree, that every account balance is non-negative and that the balances sum to a figure fully covered by the proven assets, without revealing any single balance. The Merkle branch still shows you that your account was included; the zero-knowledge proof shows that the arithmetic across the entire customer base holds. OKX runs a comparable scheme using zk-STARKs. The November proof Binance pushed out after the hack used exactly this combination, and co-chief executive Yi He noted that generating it takes about 40 to 50 hours after the snapshot is taken, a hint at how much computation sits behind a single reassuring ratio.Zero-knowledge proofs close the most embarrassing holes, but they do not move the fundamental boundary. They prove the numbers in the system are internally consistent at the instant of the snapshot. They cannot prove the list of liabilities is complete, that the keys behind the assets are still controlled, or that nothing will move the moment the camera turns off. The cryptography is honest about a narrow claim; the marketing around it often is not.

How the big exchanges prove reserves in 2026

Most large venues now run some form of proof of reserves, and the differences are in the details: whether an independent firm checks the work, how often the snapshot refreshes, and whether a zero-knowledge layer sits on top of the Merkle tree. By mid-2026, public dashboards such as the one maintained by CoinMarketCap tracked roughly $190 billion in self-reported exchange reserves, with Binance alone accounting for about two-thirds of the total, a concentration worth remembering when one venue’s proof dominates the picture.

Exchange	Core method	Independent check	Cadence
Binance	Merkle tree plus zk-SNARK	Self-run, cryptographic	Monthly
Kraken	Merkle tree	The Network Firm (accountant)	Periodic
OKX	Merkle tree plus zk-STARK	Hacken	Monthly
Bybit	Merkle tree	Hacken	Monthly
Bitget	Merkle tree	None (self-published)	Monthly
Coinbase	Audited financials, SOC reports	Deloitte (SOC 1 and 2 Type II)	Ongoing

The table flattens a lot of nuance. A self-published Merkle tree with no outside review, as Bitget has historically run, is weaker than the same tree checked by an accountant, even though both produce a tidy green ratio. And none of these methods says anything about the custody model underneath. Coinbase is the odd one out: it barely uses the phrase proof of reserves at all, because as a qualified custodian for most US spot ETFs it leans on audited financial statements and controls reports instead of a live wallet dashboard.

What a proof of reserves still cannot catch

Start with the gap that sank FTX. A proof of reserves shows assets against the liabilities the exchange reports. It says nothing about debts kept off that ledger: money borrowed from a sister trading firm, customer coins lent out and not yet returned, obligations parked in an entity that never appears in the tree. If the liability list is incomplete, a flawless proof of solvency over that list is worthless.Then there is the borrowed-for-a-day trick. Because the proof is a snapshot, an exchange short on assets can borrow coins the night before, pass the check, and return them after. Several firms were caught running versions of this in 2022, and nothing in the cryptography prevents it. The honest fixes are more frequent snapshots and unpredictable timing, not cleverer math.The Binance hack is the cleanest illustration of the snapshot limit. A proof dated 31 October cannot stop coins from leaving on some later day, and a hot wallet that drains in minutes will not wait for the next attestation. What a proof can do is answer the question afterward: once the exchange commits its own capital to replace the stolen funds, can it still show customers are covered? That is the narrow but real job the November proof was doing. Binance said the proof made the backed portion of its pledge publicly verifiable, adding that it “will absorb the financial impact of the stolen funds, not our users.” The proof did not prevent the theft or reverse it; it demonstrated the hole had been filled from the house’s pocket. A proof of reserves is a photograph, not a seatbelt. To understand how the coins left in the first place, the social-engineering playbook that opened the door belongs to the genre we covered in the bait-and-switch bots stalking crypto workers.

The Liquid lesson: when the backing itself is a lie

There is a subtler failure that even a complete, independently checked proof of assets would miss, and 2026 produced a textbook case. On 6 September, an attacker drained about $320 million in Bitcoin from the Liquid Network, the sidechain built by Blockstream. Chainalysis traced the exploit to a caching bug in the proof-validation code of Elements, the software Liquid runs. The flaw let the attacker mint roughly 4,000 L-BTC that were backed by no real Bitcoin, then use the ordinary peg-out process to redeem the fake tokens for genuine coins.Sit with what that means for reserve proofs. On chain, the L-BTC supply looked fully backed. The tokens existed, the ledger accepted them, and any naive dashboard would have shown reserves matching liabilities. The backing was a lie not because someone cooked a spreadsheet, but because the code that was supposed to enforce backing had a hole in it. A proof of assets that counts tokens the system itself treats as valid will faithfully count counterfeit ones too. The attackers, who called themselves white hats, eventually returned about 85 percent and kept roughly $47 million, which does not soften the lesson: verifiability has to reach down to the logic that mints and backs an asset, not just the balance sitting on top of it.This is the strongest argument for assets whose backing you can inspect directly. Bitcoin held at a plain address is backed by nothing but itself, and anyone can check the balance in seconds. The moment a claim on Bitcoin passes through a wrapper, a sidechain, or an exchange’s internal ledger, you are trusting that the machinery translating the claim into real coins works exactly as advertised. Most of the time it does. Liquid is the reminder of what it costs on the day it does not.

FTX and the recovery that was not really a recovery

Proof of reserves exists because of specific disasters. The collapse of Mt. Gox in 2014 lost hundreds of thousands of customer bitcoin and taught the first generation of holders that an exchange balance is a promise, not possession. FTX in 2022 taught it again at far greater scale, and with a twist that still shapes how people read recoveries today.The FTX estate has in fact paid creditors back. By the fifth distribution in mid-2026, the FTX Recovery Trust had sent out more than $10 billion, with the smallest claims recovering 120 percent and most customer claims clearing 100 percent of their filed value. On paper that is a remarkable outcome for a fraud of that size. In practice many customers feel robbed, because claims were fixed in dollars at the November 2022 petition date, when Bitcoin sat near $16,000. A customer owed one Bitcoin received the 2022 dollar value of that coin, not the coin, and then watched it recover to a price several times higher without them. Getting your money back is not the same as getting your coins back.That distinction is the whole case for treating reserve proofs as evidence rather than comfort. An exchange that can prove solvency today is far safer than one that cannot. But even a solvent, well-run exchange converts your claim on a specific asset into a claim on a company, and companies fail, freeze withdrawals, and get settled by bankruptcy judges in the currency of the day. The maxim that your keys mean your coins is not ideology; it is a description of who bears which risk.

An attestation is not an audit

Most exchange proofs that carry an outside name attach an attestation, not an audit, and the difference is larger than it sounds. In an attestation, often an agreed-upon-procedures engagement, the exchange defines the specific checks and the accountant confirms those exact checks passed at a point in time. The firm is not expressing an opinion on the business as a whole, is not hunting for undisclosed liabilities, and is not vouching that the scope covered everything that matters. It is confirming that the steps it was asked to run produced the stated result.The limits showed up almost at once after FTX. When a major accounting firm produced an early reserve report for Binance in December 2022, it drew heavy criticism for how narrow the scope was, and the firm soon paused crypto reserve work for its exchange clients altogether. The current roster, firms such as The Network Firm for Kraken and Hacken for OKX and Bybit, does careful work within defined limits, but a reader has to understand where those limits fall. A green ratio signed by an accountant means the agreed checks passed, not that an auditor stands behind the solvency of the entire company.Coinbase sits at the other end of that spectrum and shows the tradeoff. It publishes SOC 1 and SOC 2 Type II reports from Deloitte and files audited financial statements as a public company, a stronger form of assurance than any point-in-time attestation. But it declines to publish a live, wallet-level proof of reserves. Chief financial officer Alesia Haas has said the company would never disclose the addresses it holds on behalf of clients, citing security, and asks customers to rely on audits instead. More assurance, less live transparency: you cannot push both to the maximum at once.

The United States has no rule, and one loud exception

Here is the fact that surprises most American readers: no US law requires a crypto exchange to publish a proof of reserves. The practice is entirely voluntary, driven by competition and the memory of FTX rather than by the Securities and Exchange Commission. An exchange operating in the United States can run a full, independently checked proof of reserves, a bare self-published Merkle tree, or nothing at all, and stay within the rules either way.What the US does regulate is custody around the edges. The SEC’s qualified-custodian framework governs how registered investment advisers hold client crypto, which is why ETF issuers route coins to names like Coinbase and BitGo. A 2025 accounting change mattered here too: the SEC rescinded the rule known as SAB 121, which had forced custodians to book safeguarded customer crypto as a balance-sheet liability and had effectively kept banks out of the business. And the protections ordinary savers assume do not apply here: neither SIPC nor FDIC insurance covers crypto holdings, because a token is neither a bank deposit nor a registered security.The one loud exception is stablecoins. The GENIUS Act, signed in July 2025, requires every payment stablecoin to be backed at least one to one by cash, insured deposits, and short-dated Treasuries, and it forces issuers to publish a monthly breakdown of those reserves examined by an accounting firm registered with the PCAOB, with full annual audits once an issuer passes $50 billion in coins outstanding. As the Richmond Fed summarized it, that is a genuine, enforceable reserve-disclosure regime, and it lands on the dollar tokens traders park between positions, not on the exchanges holding them. The United States mandates reserve transparency for the stablecoin and leaves it optional for the venue.

Self-custody: you are your own proof of reserves

There is one custody model where the whole debate evaporates. If you hold your own keys, your reserve is a public address and you verify it yourself, every block, with no attestation, no accountant, and no faith in a snapshot. Self-custody collapses proof of assets, proof of liabilities, and proof of solvency into a single fact you control: the coins are at an address, you hold the key, and the balance is right there on chain. The same is true of well-run multisig and collaborative-custody setups, where the spending rules live in public code you can read.The catch is that verifiability and legal recourse run in opposite directions. The more of the proof you can check yourself, the less anyone else is obligated to make you whole. Lose your seed phrase and no proof of reserves, no court, and no insurer brings the coins back. Hand the coins to a qualified custodian or a bank and you gain segregation, audits, and someone to sue, while giving up the ability to see the reserve with your own eyes. On-chain venues sit in an interesting middle: a decentralized exchange like the one we examined in our piece on Hyperliquid’s yield and buyback math publishes its balances and flows on chain by design, so reserves are transparent by construction rather than by a monthly report. The table below maps the trade.

Where your coins sit	Verify the reserve yourself?	What you must trust	Recourse if it vanishes
Self-custody (hardware wallet, seed)	Yes, fully, on chain	Only yourself	None
Multisig or collaborative custody	Yes, rules are public code	Co-signers and the setup	Limited, by contract
Exchange with proof of reserves	Partly: your leaf and the total	Completeness and timing of the proof	Unsecured creditor in bankruptcy
Exchange without proof of reserves	No	The company’s word	Unsecured creditor in bankruptcy
Qualified custodian (SEC framework)	No, through audits only	Auditors and regulators	Segregated assets, legal claim
Spot ETF wrapper	No	Issuer and its custodian	Regulated fund structure

No row in that table is the right answer for everyone. A trader who needs deep liquidity and instant withdrawals lives on exchanges and should treat a strong proof of reserves as a real, if partial, signal of safety. A long-term holder who values nothing above control should be able to verify a balance without asking anyone’s permission. Most people end up spread across several rows, which is fine, as long as they know which risk each row carries.

How to actually read a proof of reserves

Treat a proof of reserves as a document to interrogate, not a badge to admire. A handful of questions separate a meaningful proof from a logo on a landing page.

• Does it pair assets with liabilities? A page that shows only how many coins the exchange holds, with no total of what it owes, is proof of assets dressed up as proof of solvency.
• Who checked it, and how? A self-published Merkle tree is a starting point; an attestation from a named accounting firm is stronger; full audited financials are stronger still. Know which one you are looking at.
• How fresh is the snapshot? A proof from last quarter says little about today. Monthly or more often, with unpredictable timing, is far harder to game than a scheduled annual photo.
• Can you verify your own balance? A real Merkle proof lets you confirm your account was included. If there is no way to check your own leaf, the liabilities side is taking the exchange at its word.
• Does a ratio above 100 percent lull you? Solvency at a snapshot is not safety over time. It says nothing about tomorrow’s hot-wallet breach, a frozen withdrawal queue, or a liability hidden off the ledger.

The same instinct applies to custody choices more broadly. Friction is often a feature rather than a flaw: the steps that make self-custody feel slow are the same steps that put the reserve under your own eyes, a point we argued at length in our case for deliberate friction in wallet design. A proof of reserves is worth reading precisely because it forces the exchange to show its work, and forces you to decide whether the work is enough.

What a better proof would look like

The gaps are not permanent. The clearest upgrade is continuous proof instead of a monthly photograph: an exchange that commits balances and liabilities on a rolling basis, so the window an attacker or a bookkeeper could exploit shrinks from thirty days to minutes. Several teams are building toward on-chain verification too, where the zero-knowledge proof is checked by a smart contract rather than trusted because a company published it, which would let anyone confirm solvency without taking the exchange’s dashboard on faith. Vitalik Buterin argued in the same post-FTX essay that exchanges should move in exactly this direction, toward proofs the public can verify rather than attestations the public must believe.The harder problem is standardizing the liabilities side. Proof of assets is close to solved; proof of liabilities still depends on the exchange honestly enumerating what it owes, and there is no shared format, no mandatory inclusion of every account, and no cross-checking against off-ledger debt. A liabilities standard that forced every customer obligation into the tree, verified by an outside party with a real audit mandate, would close most of the room for games. Until then, the strongest proofs pair a cryptographic liabilities tree with a financial-statement audit, and even that combination is rare.One limit no proof can erase is worth naming directly. The Binance breach was not a solvency failure; the exchange was solvent before the theft and stayed solvent after, because it absorbed the loss. It was an operational-security failure, a compromise of the people and machines that control the keys. No proof of reserves, however continuous or cryptographic, measures how well an exchange guards its hot wallets or screens its job applicants. Solvency and security are different questions, and a flawless answer to one tells you nothing about the other.

The bottom line

Proof of reserves grew up in 2026. The cryptography is better, zero-knowledge proofs are standard at the largest venues, and an exchange that had just lost $6.3 billion could still show its customers were backed within days. That is real progress, and it is worth rewarding exchanges that do it well. But the phrase promises more than any single proof delivers. A proof of assets is not a proof of solvency. A snapshot is not a guarantee about tomorrow. An attestation is not an audit. And an on-chain balance that passes every check can still be counterfeit if the code minting it is broken, as Liquid showed.Read the proofs, reward the transparent venues, and keep in view what sits at the far end of the spectrum. The only reserve you never have to take on faith is the one whose key is in your own hands. Everything else is a question of how good the evidence is, and how much recourse you keep if the evidence turns out to be wrong.

Frequently Asked Questions

What is proof of reserves in crypto?

Proof of reserves is a way for a crypto exchange or custodian to show it holds enough assets to cover what it owes customers. A full version pairs proof of assets, meaning evidence the firm controls the coins, with proof of liabilities, meaning the total it owes, so the two can be compared. Most exchanges in 2026 build it from a Merkle tree, often with a zero-knowledge proof on top.

Does proof of reserves mean my money is safe?

Not by itself. A proof of reserves shows solvency at the moment of a snapshot, and only over the liabilities the exchange chose to include. It cannot catch hidden debts, coins borrowed just for the check, or a hack that drains the wallet the next day. It is a useful signal, not a guarantee.

Which crypto exchanges have proof of reserves in 2026?

Most large venues do, including Binance (Merkle tree plus zk-SNARK), Kraken (Merkle tree reviewed by The Network Firm), OKX and Bybit (Merkle trees checked by Hacken), and Bitget (self-published). Coinbase relies instead on audited financial statements and SOC reports as a qualified custodian.

Is proof of reserves required by law in the United States?

No. US law does not require exchanges to publish a proof of reserves, so the practice is voluntary. The main mandatory reserve-disclosure rule is the GENIUS Act, which makes payment stablecoin issuers back their tokens one to one and publish monthly reserve reports examined by a registered accounting firm.

What is the difference between proof of assets and proof of solvency?

Proof of assets shows a firm controls a certain amount of crypto. Proof of solvency shows those assets are at least as large as everything the firm owes customers. Proof of assets alone can mislead, because an exchange can hold large reserves and still owe far more than it holds.By Marcus Okafor, senior markets writer at HOGE Wire, covering custody, exchanges, and market structure.

More from this desk

Wallet UX in 2026: The Case for Friction

Nov 3, 2026

Ledger vs Trezor vs Tangem: The Seed Phrase Is the Leak

Oct 30, 2026

Nobody Cracked a Key: Crypto’s October 2026 Wallet Wave

Oct 29, 2026