Binance Lost $6.4 Billion in Twenty Minutes. Here Is How North Korea Launders It, and Why Only Half Could Be Frozen

2026.10.31 Author: Robert Nogacki

On the afternoon of October 29, 2026, a few minutes after four o’clock in New York, which was a few minutes past nine in the evening in Warsaw, roughly $6.4 billion began to drain out of the hot wallets of Binance, the largest cryptocurrency exchange in the world. Most of it was gone within twenty minutes, from eleven blockchains at once. Three firms that make a business of following money through public ledgers, Chainalysis, TRM Labs, and Elliptic, point to TraderTraitor, the North Korean unit that the F.B.I. blamed for the $1.5-billion Bybit theft of February, 2025, although each says so with a different degree of confidence, and no government has yet said so at all. Within hours, somewhere between $2.7 billion and $2.85 billion had been frozen, almost all of it stablecoins, or about $3 billion by Binance’s count. More than $3 billion in Bitcoin, Ether, XRP, and TRX remained in the thief’s hands, and no one on earth has a button that would stop it. What follows is the trail of that money, hour by hour, and then an explanation of why “freezing crypto” is at once the fastest and the least reliable instrument the law possesses. The facts are as reported in public analyses and statements through October 31, 2026; where something is the exchange’s own account, an analytics firm’s assessment, or a press report, I say so.

Thirty-Six Hours Before the Theft

On the morning of October 28th, three small sums landed in a wallet on the Ethereum network. They came from Tornado Cash, a mixer that pools the coins of many users to sever the link between what goes in and what comes out. From that wallet, the next day, the operator deployed a first version of the attack code and tried it on two test wallets, rehearsed the final version on a third in the afternoon, and sent the test funds back to himself. Four hours after the dress rehearsal, the same code was sitting on a Binance hot wallet. That is the reconstruction by TRM Labs, which traced every transfer from the exchange’s wallets to the attacker’s addresses; Elliptic describes the same preparations.

A hot wallet is a wallet connected to the Internet, from which an exchange pays out customer withdrawals; unlike cold wallets, which are kept offline, it has to sign transactions around the clock. This particular Binance wallet on Ethereum is among the best-known addresses in the industry.

The weapon was a convenience feature. Since the Pectra upgrade of May, 2025, an ordinary Ethereum wallet can, by signing a single authorization under a standard called EIP-7702, hand control of its behavior to a smart contract: from then on, every call to the wallet runs that contract’s code, until a new authorization replaces it. At 4:20 P.M., the operator submitted one transaction containing fifty such authorizations. Each was signed by the Binance hot wallet, and each pointed it to a “sweeper” contract that would take orders only from the attacker. An authorization works only at one precise point in the wallet’s running count of transactions, and the Binance wallet handles a ceaseless stream of withdrawals, so that count keeps climbing. Fifty consecutive numbers all but guaranteed that one of them would match at the moment of execution. One was all it took.

The thief did not crack the safe. He got the safe to sign fifty blank forms and waited for one of them to come due.

How he obtained the signatures is still unknown. Binance first spoke of a compromised laptop belonging to a DevOps engineer, through which the attacker reached a service container and submitted unauthorized withdrawals. In a statement of October 30th, the exchange added that its investigation had uncovered further points of breach, that it had found no evidence that private keys had leaked, and that a technical post-mortem would follow once verified. Richard Teng, Binance’s co-chief executive, described a complex attack, the work of more than one person, involving malware that erased its own tracks and a manipulation of the transaction-approval system so that withdrawals bypassed the usual controls. TRM goes further: if, within half an hour, Binance wallets on many chains, with different key types and signing schemes, all produced valid signatures for the attacker, the simplest explanation is that the infrastructure that signs for the exchange’s hot wallets was itself taken over. Elliptic offers a comparison worth keeping: Bybit went the same way, when a developer’s machine at the wallet provider was compromised and the transaction that Bybit’s signers approved in good faith was swapped out beneath them. In both cases, what broke was not the cryptography but the people and the plumbing around it. The exchange’s version is a statement; the analysts’ version is an inference from the ledger; the truth sits in logs that no outsider has seen.

Twenty Minutes, Eleven Chains

The first move came where no one was looking: the XRP Ledger. Between 4:05 and 4:09 P.M., five Binance wallets sent 233.46 million XRP, worth about $296 million, to a single attacker address, leaving each wallet with only the minimum balance the network requires. That XRP has not moved since. Over the next thirty minutes, wallets were emptied on Solana, NEAR, Base, Optimism, Bitcoin, Tron, Ethereum, Arbitrum, and Avalanche. At 4:20, a Bitcoin hot wallet sent 18,270 BTC, about $1.47 billion, to an address that, within the same block, began slicing it into tidy pieces of 1,000 and 100 BTC. The technique is called a peel chain, and it does not hide money so much as multiply tracks: instead of one sum, investigators get a dozen or more, each with a life of its own. Within two minutes, a hot wallet on Tron gave up 612.7 million USDT and 46.4 million TRX.

On Ethereum, three minutes and twenty-four seconds after the authorizations went in, the operator ordered the Binance wallet to send its entire balance to his address. A single transaction, visible to anyone in a public explorer at 4:24 P.M., moved 167,464 ETH and sixty-three kinds of tokens, among them 1.615 billion USDT; by Elliptic’s count, that was two-thirds of all the Ethereum losses, $2.53 billion in one transfer. The authorization stayed live, so as customers kept depositing into the hot wallet the attacker swept it again, at 4:41, 4:59, 5:04, 5:44, 6:39, and 7:56. The same wallet served as the executor of a separate Binance smart account, which let the attacker pull 823 million USDC out of it. Another Binance hot wallet on Ethereum sent 28,006 ETH, 55.7 million USDT, and 25.9 million RLUSD by ordinary signed transfers, with no delegation at all; Elliptic puts such transfers at $174 million. A second wave, a little after seven in the evening, took wallets on Avalanche and Polygon; Aave’s risk team logged 444,537 AVAX landing on the attacker’s address at 7:04 P.M., proof that the compromise was still live more than two hours after the first alarm.

TRM’s tally, at prices at the moment of theft: Ethereum, about $3.79 billion (200,630 ETH and a hundred and eighty-five other tokens, including 1.875 billion USDT, 823 million USDC, and 88.5 million RLUSD); Bitcoin, 18,270 BTC; Tron, about $628 million; the XRP Ledger, about $296 million; Solana, 970,500 SOL, about $96 million; Arbitrum, Optimism, and Base together, about $39 million; Avalanche, about $28 million; Polygon, about $10 million. Altogether, about $6.36 billion across ten chains. Elliptic counts $6.43 billion and eleven chains, because it adds NEAR, and notes that four assets, USDT on Ethereum, Bitcoin, USDC, and USDT on Tron, made up seventy-four per cent of the haul. Chainalysis says $6.4 billion; Binance says $6.3 billion. The differences come from prices, timing, and what each firm counts, not from any dispute about what happened. Elliptic also flags something easy to miss in the rush: after Binance halted withdrawals, a further $169 million left the exchange for old, well-established addresses with no link to the attacker, which looks like Binance moving its own money, not another theft. Not every large transfer after the alarm is loot.

For scale: Elliptic judges this almost certainly the largest theft of any kind in history. The previous record, before Bybit broke it, belonged to Saddam Hussein, who removed a billion dollars from the Central Bank of Iraq on the eve of the 2003 invasion. This was more than six times that, and it required no trucks.

The Race Against the Freezer

From that moment, the attacker was not running from the police. He was running from the stablecoin issuers. Bitcoin and Ether have no issuer, so there is no one to write to; USDT, USDC, RLUSD, and USD1 do, and each issuer has written into its token’s contract a blacklist function that freezes the balance of any address it names. Stablecoins, by Elliptic’s count, made up $3.46 billion of the haul. The thief’s priority, then, was to convert what could be frozen into what could not, before anyone reached for the button.

According to Chainalysis, the attacker moved about $500 million in stablecoins into a wallet whose sole job was conversion and, in under twenty minutes, sold about 325 million USDT for about 122,300 ETH. TRM did the arithmetic: that worked out to nearly $2,650 per Ether while the market was paying $2,300 to $2,350. Elliptic, measuring a wider window, sees 434 million USDT swapped for 160,451 ETH through Uniswap, 1inch, and the request-for-quote systems of market makers, at a loss of twelve to twenty-six per cent against the market price. The two figures must not be added, since they describe the same operation at different moments, and both say the same thing: the thief paid the market a double-digit premium for speed. The remaining 175 million USDT from the first batch was frozen before it reached an exchange.

The most instructive episode required no mixer. Between 4:58 and 5:01 P.M., a wallet in the attacker’s cluster deposited 60 million USDT as collateral in the lending protocol Aave and borrowed 19,762 ETH against it, worth about $44.6 million at the time. That is the account of Aave’s risk team, which, within ten minutes of the first reports, convened a war room and decided to freeze the protocol’s markets on every network: a hundred and seventy-five reserves on Aave V3 across sixteen networks and a hundred and seventeen on Aave V4. The first freezing transaction was signed at 5:42, forty-one minutes after the loan. But a freeze on Aave means something different from a freeze at an issuer: it blocks new deposits and new loans, while repayments, withdrawals, and liquidations go on. The loan was already out the door. The attacker’s position, $59.9 million of collateral against $43.8 million of debt, is over-collateralized and subject to the ordinary rules of liquidation, so the risk team judged that the protocol’s lenders were not at risk; by the thirty-first, Elliptic described the loan as still open and close to liquidation. What the lenders had lost was certainty about who owns the 60 million USDT now sitting in their pool. On Tron, 300 million USDT went to a wallet that began lending it into a protocol called JustLend, and 234 million USDT was swapped on SunSwap for 638 million TRX. Another 200 million USDT passed through a chain of intermediary wallets and a swap on Uniswap, of which 30 million was caught and frozen. Of 88.5 million RLUSD, the attacker sold 48 million on decentralized exchanges between 5:42 and 6:01, before Ripple froze the rest. Some hundred and fifty other tokens, UNI, AAVE, LINK, and the rest, were dumped for Ether as they came. Elliptic estimates that, before the freezes landed, about $786 million of stablecoins had been turned into Ether and TRX.

It is the most expensive laundry in the world: here, every laundered dollar cost a dime or two in discounts to the market and left a complete record behind. For a criminal who has to answer to an accountant, that would be a catastrophe. For a state that steals to fund a weapons program, it is an operating cost.

Who Froze What, When, and on Whose Authority

The response was, by the standard of every previous theft, fast. Tether froze 971.1 million USDT in the main receiving wallet at 5:01 P.M., thirty-seven minutes after the main sweep, in the same minute that the attacker was finishing his loan on Aave. Later sweeps landed in a wallet that was already frozen, which now holds 1.175 billion USDT; Tether also blocked 205 million in two other Ethereum wallets, 378.7 million in two wallets on Tron, and smaller sums on Avalanche and Polygon, about $1.77 billion in all, along with some tokenized gold. Circle froze 823 million USDC at 5:04, in coördination with law enforcement, by its own account, and later added 1.56 million that the attacker had managed to swap out of AVAX and bridge over to Ethereum. Ripple, at 6:01, locked the remaining 40.5 million RLUSD and froze RLUSD on the compromised Binance wallet as well, so that fresh deposits could not be swept. World Liberty Financial froze 27.8 million USD1 at 6:10; Paxos and First Digital blocked their own stablecoins, and Binance froze WBETH, a token it issues itself. Finally, at 8:08 that evening, the Security Council of the Arbitrum network moved 8,425 ETH and 74,559,805.2 ARB, about $30 million, out of the attacker’s reach, and two days later the governance of Optimism appears to have done the same with 44 million OP, about $5 million. TRM counts about $2.7 billion in all; Elliptic, as of the morning of October 31st, $2.85 billion frozen or seized, forty-four per cent of the theft, ninety-four per cent of it stablecoins; Binance says about $3 billion. Arbitrum and Optimism are the only networks on which everything stolen was secured, in both cases by the network’s own governance. On Bitcoin, the XRP Ledger, Solana, and NEAR, nothing was frozen, because there was no one to do it.

How the Arbitrum council did it deserves a closer look, because it is a lesson in what “decentralization” actually means. According to the council’s official account, it carried out a single atomic operation: it temporarily replaced the network’s Inbox contract on Ethereum with a version containing a function that can send a transaction on behalf of any address, used that function to execute two transfers in the attacker’s name, moving his funds to an address controlled by the DAO, and then restored the previous contract. In other words, a group of a dozen or so people holding keys can, on this network, make it execute a transfer from anyone’s address without that person’s signature. This time, they used that power against a thief. The frozen funds can now be moved only by a vote of ArbitrumDAO’s token holders.

Here the legal part begins, because there is not a single court on that list, or none that has been made public, and no prosecutor, regulator, or law-enforcement agency whose order anyone has seen. There are private entities exercising their own contractual powers: issuers using a function written into the code of their tokens and into their terms of service; a lending protocol using its guardian’s authority; the Arbitrum council using powers set out in the network’s governance rules. The United States Secret Service said that it was working around the clock with private-sector partners, including Circle, to identify, freeze, trace, and recover the funds, and the regulator in Abu Dhabi that issued Binance’s main license said that it was monitoring the incident. The state was present, then, as a partner and an observer. The button was pressed by companies.

It is also worth being precise about what a freeze is and is not. The frozen tokens, except those the two networks carried off, still sit at the thief’s address. The blockchain displays them to anyone who cares to look. They simply cannot be moved. For them to return to Binance, or to anyone else, the issuer typically has to destroy them and mint them afresh to a different address, or release the dollars behind them, which it does under its own terms of service, usually on a demand from an authority or a court order. Binance itself, in its statement, separates “frozen” from “recovered”: recovery it describes as work still ahead of it. A freeze, then, is a pause, not a resolution. The question of whose money this is has only now reached the lawyers.

What Cannot Be Frozen

After the freezes, the attacker controls, by Elliptic’s count, about $3.21 billion in assets that no issuer can block, scattered across dozens of addresses: 519,104 ETH, 18,270 BTC, 233 million XRP, and some 475 million TRX. Bitcoin has no issuer; Ether has no issuer; XRP on the XRP Ledger has no issuer who could put an address on a blacklist, and Ripple can freeze RLUSD but not XRP. These assets can be followed; they cannot be stopped. The Bitcoin was split through a short peel chain into nineteen addresses, most holding exactly 1,000 BTC, and none of them has yet touched any service. The XRP, SOL, and NEAR, together about $437 million, have not stirred since the theft. Everything that remains now depends on whether the thief will one day have to touch a place where someone asks for identification.

Chainalysis is watching laundering patterns it says it has never seen before: thousands of transactions in a short span, funds cycling among addresses and returning to addresses used earlier in the sequence, and a repeating signature in which an address first receives exactly 1 TRX, then the real money, then passes it on. That is the signature of a script, not a hand. The numbers bear it out. On the public list of addresses that Elliptic released for screening, a snapshot taken early on October 31st held 2,847 indicators, and the list has grown since; by my own count of that file, about seventeen hundred of them are Tron addresses, about eleven hundred are addresses on Ethereum-compatible networks, and only a few dozen are Bitcoin. The machine works mostly on Tron, where transactions are cheap.

The precedent is fresh. On February 21, 2025, the same group took about $1.5 billion in Ether out of Bybit. In ten days, all of it had been moved on and converted, mostly into Bitcoin through THORChain, a cross-chain swap protocol, whose volume swelled to nearly $5.9 billion in the process, with about $5.5 million in fees; one of its developers quit in protest. Bybit’s chief executive summed it up on March 4, 2025: seventy-seven per cent of the funds could still be traced, twenty per cent had gone dark, and three per cent had been frozen. Three per cent. After the loot was converted into Bitcoin and scattered across more than nine thousand wallets, that share barely rose. Elliptic notes that it labelled the first Bybit addresses within eighteen minutes, and the laundering still took months.

The road ahead is known from earlier cases: cross-chain bridges, instant-swap services that ask no questions, over-the-counter brokers, and, increasingly, Chinese money-laundering networks that turn crypto into cash. I have written about that infrastructure at length (The Shadow States: How Iran, Russia, and North Korea Turned Cryptocurrency Into the Infrastructure of Sanctions Evasion). There is also a second, less obvious move: do nothing. A thief who is chased by neither a statute of limitations nor a creditor can wait for the world to busy itself with something else.

And Tornado Cash, where it all began? In August, 2022, the U.S. Treasury placed it under sanctions; in November, 2024, a federal appeals court held that immutable code cannot be treated as “property” subject to sanctions; on March 21, 2025, the mixer was removed from the sanctions list. Its co-founder Roman Storm was convicted in August, 2025, solely of conspiring to run an unlicensed money-transmitting business; the jury could not agree on the money-laundering and sanctions counts, and a retrial on those two counts is now set for April, 2027. The tool with which the largest heist in the history of cryptocurrency was prepared is not, today, under American sanctions. That is not a complaint about the courts; it is a description of the law’s limits. You cannot arrest code, and it is hard to convict its author for what other people did with it.

Who Did It: Three Firms, Three Degrees of Certainty

The strongest public evidence arrived in real time. Funds from Binance commingled with funds stolen on September 24th from the exchange Bitget, $387.5 million that Chainalysis and Elliptic have attributed to North Korea. In its first update, Elliptic retraced the route: from an attacker address on Tron through a peel chain of about twenty-eight hops, then through THORChain and the Hyperlane bridge to Ethereum, to an address that had earlier received Bitget money; the investigator known as ZachXBT spotted it first. On that basis, Chainalysis attributed the attack to TraderTraitor, and TRM speaks of high confidence, grounded also in intelligence from off the chain.

Elliptic is more careful. It writes that North Korean involvement is likely, not certain, and the reservations it has set out across its updates deserve attention, because they are the best available test of how to read evidence on a ledger. First, the overlap is small: about $39,000 from Binance has reached the shared address so far, against about $510,000 from Bitget, and a professional laundering network can serve more than one client; a shared laundry does not prove a shared burglar. Second, the tempo: Pyongyang’s launderers usually swap stablecoins into native assets within minutes, yet here more than $2.6 billion in stablecoins was still sitting in the attacker’s wallets when the freezes came down; the USDT was held for thirty-seven minutes on Ethereum and forty-six on Tron. Third, the method: market makers and lending protocols rather than bridges, as after Bybit. Fourth, the most diagnostic fact is still missing, which is how the engineer’s computer was compromised. And fifth, the F.B.I. attributed Bybit to North Korea within five days; here it has said only that it is aware of the incident and is coördinating with partners. An honest account can describe a strong intelligence assessment. It should not promote it to a finding.

Why the Law Loses to the Clock

Set three clocks side by side. The thief: about twenty minutes to move most of $6 billion off eleven chains. The issuers: thirty-seven minutes to five hours to freeze. And Polish law, which, had the same money passed through an obligated institution in Poland, provides the path set out in Article 86 of the Act on Counteracting Money Laundering and Terrorism Financing (ustawa o przeciwdziałaniu praniu pieniędzy oraz finansowaniu terroryzmu): the institution notifies the General Inspector of Financial Information (Generalny Inspektor Informacji Finansowej) at once and may hold the flagged transaction for up to twenty-four hours; the Inspector may order the transaction suspended or the account blocked for up to ninety-six hours; a prosecutor may then block the account by order for up to six months, and for a further period of up to six months after that; and the block lapses if, in that time, no order is issued securing the assets or treating them as material evidence. Article 92 of the same act excludes Saturdays and public holidays from the running of the hourly limits. None of these is a waiting period; each is a leash of fixed length, and at the end of each a different authority has to take hold of the next one. The institution gets a day, the Inspector four working days, the prosecutor half a year, and all of it presupposes that someone in Poland noticed the transfer and reached a party able to stop it. Tether’s freeze took thirty-seven minutes and needs no one’s renewal. I have described the procedure from the other side of the counter in an earlier piece (When the State Freezes Your Money).

The instrument, moreover, was cut for a bank. When, in 2021, the Supreme Court (Sąd Najwyższy) held in two resolutions that funds in a bank account do not exist as things and lack the attributes of material evidence, the legislature added Article 236b to the Code of Criminal Procedure (Kodeks postępowania karnego), under which, for the purposes of the rules on seizing things, funds in an account are things as well. The anti-money-laundering act has, since 2018, counted among “accounts” an electronic record that gives its holder access to units of virtual currency, so a block under Article 86 reaches an account at a Polish provider; the Code of Criminal Procedure knows no such definition, and the Criminal Code’s definition of movable property, in Article 115 § 9, covers money, other means of payment, and funds recorded in an account, and whether a virtual currency is an “other means of payment” is still argued. In practice, a prosecutor has three roads: a block at a Polish crypto-service provider, which has been an obligated institution since 2018; a securing of assets under Article 291 of the Code of Criminal Procedure, executed the way a property right is attached in civil enforcement; and the physical capture of a private key when things are seized, after which the funds are moved to an address the authority controls. The third road works only if the key is within reach of a Polish officer, and a seized computer is not yet seized money: as long as someone else can still sign a transfer, a device in an evidence locker and funds under an authority’s control are two different things. In the Binance case, the keys are wherever the attacker is, which, if the analysts are right, is Pyongyang.

Abroad, things are slower, not faster. Within the European Union, Regulation 2018/1805 on the mutual recognition of freezing and confiscation orders allows, in urgent cases, up to forty-eight hours to recognize an order and another forty-eight to execute it; outside the Union, there is mutual legal assistance, measured in weeks and months, and when the perpetrator is a sanctioned state there is not even anyone to address a request to. The E.U.’s travel rule, under Regulation 2023/1113, applicable since December 30, 2024, requires providers, for transfers of more than a thousand euros to or from a self-hosted wallet, to check whether that wallet is owned or controlled by their customer, but it bites only at the point of contact with a regulated intermediary; it gives the state no key to a self-custodied wallet. What exchanges must report about their customers in any event, I have described separately (Cryptocurrency Transaction Reporting). The Union has noticed the problem. Directive 2024/1260 on asset recovery and confiscation, which member states must transpose by November 23, 2026, writes crypto-assets expressly into the definition of property and requires that asset-recovery offices be able to freeze immediately, for up to seven working days, in cross-border cases where there is an imminent risk that property will disappear and no other authority can act in time. That is a step in the right direction. The question remains what “immediately” means inside an apparatus that has so far measured time in days, and who in Poland, on a Thursday night, would send Tether a demand that Tether would accept as sufficient.

Because in this case the role of asset-recovery office was played by a company headquartered in El Salvador, which acted after thirty-seven minutes, on its own responsibility and under its own terms of service. One can be glad of that. One can also ask what happens on the day the same company freezes the wrong thing, or fails to freeze the right one, because no statute tells it when it must act and when it must stand back; its own terms of service do, and it wrote them itself.

Whose Tokens Are They?

A freeze stops the money, but it does not answer the question of whose money it is, and that question turns out to be harder with crypto-assets than with a bank. USDT is as fungible as cash: one token is indistinguishable from the next. When the thief sells hundreds of millions of USDT to market makers, those market makers receive stolen tokens and pay for them with real Ether. When he deposits 60 million USDT in Aave and borrows Ether against it, the lenders are left holding stolen collateral. Each of these parties will say that it acted in good faith. Each will have to explain why, at ten o’clock at night, it was buying hundreds of millions of dollars from an address that was giving away a dollar for eighty-odd cents.

Polish law has no ready answer. Crypto-assets are not things within the meaning of Article 45 of the Civil Code (Kodeks cywilny), so the rules on acquiring movable property in good faith from a non-owner, in Article 169, do not apply directly, and scholars disagree about what applies instead. The criminal law is sterner than the civil: Article 45 § 3 of the Criminal Code (Kodeks karny) directs that, in the serious cases to which extended confiscation applies, and a six-billion-dollar theft is one many times over, property constituting the proceeds of crime, transferred to a third party in fact or under any legal title, is deemed still to belong to the perpetrator, unless the circumstances of the acquisition gave no reason to suspect that it came, even indirectly, from a prohibited act. The burden of explaining the circumstances therefore falls on the acquirer; how far that mechanism reaches, I have set out elsewhere (Criminal Asset Forfeiture and Extended Confiscation). Directive 2024/1260 takes the same road: it permits confiscation from a third party who knew or ought to have known that the transfer served to avoid confiscation, and lists, among the indicators, an acquisition at a price markedly below market value. A market maker who got a dollar for eighty-odd cents will have to reckon with that indicator. On the other hand, mere technical contact with a flagged address makes no one a launderer: Article 299 of the Criminal Code requires intent, not just a line on a chart (Money Laundering under Article 299 of the Polish Criminal Code).

An English court showed in 2024 how this plays out in practice. In D’Aloia v. Persons Unknown and the exchange Bitkub, the High Court held that USDT is property that can be traced and held on trust, and then dismissed the claim against the exchange, because the plaintiff could not show that his particular tokens had reached the exchange’s wallet. The tracing method adopted by his expert was unclear, and other people’s USDT had passed through that wallet in vast quantities. The blockchain shows every transfer, but it does not say which dollar is whose. That answer has to come from a method a court will accept, and the courts are still learning it.

There is a third side to this coin, the least expected: the creditors of the state to which the theft is attributed. On April 18, 2026, attackers whom the analytics firms link to North Korea robbed a protocol called KelpDAO; three days later, the Arbitrum Security Council froze 30,765 ETH, about $71 million, at the very address where the Binance funds sit today. As the community prepared to return the money to the victims, on May 1st a law firm representing plaintiffs who hold final judgments against North Korea, in Kim v. Democratic People’s Republic of Korea and two related cases, more than $877 million awarded for acts of terrorism, served on “Arbitrum DAO” a New York restraining notice, treating the frozen Ether as North Korean property on the theory that the Lazarus Group is an instrument of the state. On May 8, 2026, Judge Margaret Garnett allowed the Ether to be transferred to Aave LLC, but the notice follows the money, and the question of who has the better right to it, the victims of the hack or the victims of the regime, the court did not decide, and nothing in the public record shows that it has decided it since. Here is the paradox of paradoxes: the better you prove that it was Pyongyang, the stronger the argument in the hands of those to whom Pyongyang has owed money for decades. Nothing suggests that such a notice has been served in the Binance case. But the mechanism is on the table, and 8,425 ETH from Binance is waiting for the same DAO vote.

And then there is the wrongful freeze. An issuer that freezes on the strength of an address list from an analytics firm can hit the address of an innocent counterparty, who has no court to go to on a Friday morning. The second presidential draft of Poland’s crypto-asset market act provides for State Treasury liability for wrongful blocks and for judicial consent to their extension, which shows that the problem has been noticed, at least on the public side. On the private side, it is governed by terms of service.

Poland: A State Without a Statute

For a Polish reader, this story has a further backdrop. The E.U.’s MiCA regulation applies across the Union, but it needs a national statute to name a supervisor, a licensing procedure, and penalties. Poland has no such statute. The President vetoed the government’s bill three times, on December 1, 2025, and on February 12th and June 11, 2026; the Sejm, the lower house of the Polish parliament, failed to override the third veto, and on September 8, 2026, a second presidential draft reached the Sejm, the one with a ninety-six-hour account block at the demand of the chairman of the Financial Supervision Authority (Komisja Nadzoru Finansowego) and judicial review of its extension. The government has announced a fourth, tougher bill. Meanwhile, the Financial Supervision Authority cannot issue a single crypto-asset-service-provider license, because it has no legal basis for one, and Polish firms are seeking authorizations in other countries.

In this arrangement, Binance has no license in the Union. After it withdrew the application it had filed with the Greek regulator, on June 24, 2026, it stopped offering regulated services to E.U. customers on July 1st; Polish users were sent withdrawal instructions, as I described at the time (Binance and the MiCA Licence: The Hostage’s Selfie). By October, the Financial Times was reporting that some E.U. customers were being served from Binance’s Abu Dhabi entity under a “reverse solicitation” exemption meant for clients who come knocking entirely on their own, and that E.U. regulators were examining the arrangement. That has a consequence few remember. Had Binance been a licensed custodian in the Union, Article 75(8) of MiCA would have made it liable to its clients by operation of law for the loss of crypto-assets through an incident attributable to it, up to their market value at the time of the loss. Without that license, its promise to make customers whole is a commercial pledge rather than the statutory guarantee MiCA would have imposed, whatever other claims a given customer may have under its contract or local law. The pledge is specific, to be fair: the exchange has set aside more than $6.34 billion, $5 billion of it in the very tokens that were stolen, is buying the rest on the market, and has promised a proof of reserves by November 2nd. For its customers, that is the most important sentence in this story. For the law, it is a reminder that, in crypto-assets, the insurer of last resort is the exchange’s balance sheet, not any guarantee scheme.

There is one more case in which speed was never the problem: Zonda, Poland’s largest crypto exchange, whose Estonian operating company was declared bankrupt in August. There, the money did not leave in twenty minutes, and the dispute is not about freezing at all: prosecutors have secured assets in Poland, and the argument is over who should return them to the victims, and under what procedure. I have written about it (Zonda: What Polish Prosecutors Can Do for the Estonian Trustee, What They Must Not Do, and a Proposal to the Minister of Justice on Paying Victims Without Waiting for a Verdict), and about the Estonian proceedings behind it (The Zondacrypto Insolvency in Estonia: Recognition Without Recovery). The paradox of the two cases is the same: it is easier to stop dollars stolen by a foreign state on a Thursday evening than to return to victims the money one’s own state has held in deposit for months.

What Follows

First, “freezing crypto” is not one institution but five, and the communiqués after the attack use one word for all of them: a blacklist in a token’s code, which immobilizes a balance but does not decide to whom it should be released; a block at a custodian, which depends on what the custodian actually controls; a protocol restriction, as at Aave, which stops new operations but reverses none already done; a network’s emergency power, as on Arbitrum, which depends on someone holding the keys and having the nerve to use them; and the state’s procedural securing of assets, the one with procedural safeguards, and the slowest. Forfeiture, or restitution to the victim, is a separate sixth step, and the Polish Criminal Code, in Article 45 § 1, keeps the two expressly apart. In the Binance case, the first four worked. The fifth is only beginning.

Second, freezability is a property of the asset and of who controls it, not of the law. The stablecoins were frozen for the most part; the Bitcoin, Ether, XRP, and SOL not at all. Any statute that promises “a block on crypto-assets” is in truth promising a block on what an intermediary holds or an issuer issues; the rest no provision will stop, only the moment, if it comes, when the thief tries to re-enter the world of documents. That moment can arrive years later: in February, 2022, the U.S. Department of Justice seized more than 94,000 BTC stolen from Bitfinex in 2016, because it found, in one defendant’s cloud storage, a file containing the private keys. No blockchain was rewound. A key was obtained.

Third, time. A procedure that measures its provisional measures in working days and half-years is a museum piece in a world where $6 billion changes hands in twenty minutes and an issuer freezes in thirty-seven. Directive 2024/1260 and the drafts of Poland’s crypto-asset market act point in the right direction, but “immediately” will become true only when a specific person is on duty, has Tether’s address, and has a signature that Tether will not question.

Fourth, private freezes need a framework. Today they work better than the state does, and today they deserve thanks for it. But a system in which the fate of billions is decided by a company’s terms of service, an address list from an analytics firm, and a vote of token holders, with no duty to act, liability waived in the fine print, and, for the wrongly frozen, an arbitration clause in a foreign forum instead of a court, is a fair-weather system. The case in New York shows what happens when the weather turns: the victims of a hack and the victims of a state line up for the same Ether, and the decision goes back to a court, only months later.

And fifth, a word to anyone reading this not for the analysis but because they have just lost money: the first twenty-four hours, the addresses, and the documents are what count. Full transaction identifiers, dates in UTC, addresses, token contracts, and amounts in the asset’s own units, kept separate from the labels an analytics vendor sticks on addresses. Then the question of who can actually stop a transfer: the issuer, the exchange the funds are flowing into, the administrator of a protocol, or the holder of a key, because a demand sent to someone without that power stops nothing. Notifying prosecutors is necessary, but it freezes nothing by itself; freezing is done by whoever has the button, and all of them move faster than registered mail. And do not let anyone tell you that a list of addresses is recovered money. A list is a valuable piece of analysis, but it is not a writ of execution, a private key, or a decision to return anything. What people bring to a lawyer in cases like these, I have described on a separate page (Cryptocurrencies).

The blockchain does not forget, but neither does it listen. It is an excellent witness, and no bailiff. The law listens, but it forgets the clock. As long as that gap endures, the great thefts of digital money will end the way this one has: with half the loot visible to everyone and untouchable by anyone.