why hackers don’t get caught

Money: a public ledger with pen names

Cryptocurrency hides names, not payments, so tracing usually ends where coins meet a business that knows its customers.

What it hides
Where payments come from and where profits go
How it gives way
A public ledger meeting an exchange that knows the customer
How much to read

Anyone who earns from crime, or pays for servers and tools, needs money that does not carry their name. Cash does not travel over the internet and bank accounts come with identity checks, so the usual choice is cryptocurrency. The trouble is that on most chains the payment trail is not hidden at all. It is public, permanent and searchable, and only the names are missing.

Pseudonymous is not anonymous

A Bitcoin address is a pen name. Nothing on the chain says who holds the key behind it, but every transaction that address takes part in is published to everyone, forever. Meiklejohn and colleagues put it in one line in 2013: ownership is implicitly anonymous, but the flow of money is globally visible 1. The analyst’s job is to turn many pen names into a few owners.

Two rules of thumb do most of that work. The first is common-input ownership. A transaction that spends coins from several addresses needs a valid signature for each of them, so in ordinary wallets all the inputs belong to one owner 1. The second is change detection. Bitcoin spends whole outputs, so a payment of 0.3 from an output of 1.0 sends the remainder back to the payer, usually to a brand-new address. Spotting which output is the change adds that address to the payer’s group.

The second rule is a habit of wallet software, not a property of the protocol, and the paper shows what happens when it is applied carelessly: a naive version fused Mt. Gox, Instawallet, BitPay and Silk Road into one “super-cluster” of 1.6 million public keys 1. With stricter conditions the authors identified 3,540,831 change addresses, at a false-positive rate they estimated at 0.17% 1. Their snapshot of the chain held 16,086,073 transactions and 12,056,684 public keys, which the two rules collapsed into 3,383,904 clusters 1. To put names on clusters they simply did business with services, 344 transactions in all, and the heuristics spread those few labels across more than 1.8 million addresses 1.

tx 1tx 2addr A0.40addr B0.35addr C0.30addr E0.20addr D0.0497addr P1.00C1: one ownerspent together: one ownerround amount: paymentodd remainder:change, joins C1spent with D: joins C1Exchangeaddr X 0.2495customer recordreal name

Transaction 2 pays a deposit address at an exchange, and the exchange knows who opened that account. One customer record names the whole cluster, including addresses that never touched the exchange.

Clustering on a public ledger. The default view shows every inference at once: common input groups A, B and C, the odd remainder marks D as change, E joins through D, and one exchange record names all five. The buttons step through the same reasoning in order.

The figure is the whole layer in small. Everything an analyst uses is readable to anyone: addresses, amounts, times, which outputs feed which inputs. The one thing missing is the owner, and that is supplied by somebody off the chain who already knows it.

Large balances rarely move in one jump. A common pattern is the peel chain: a big output pays a small amount to someone and sends the rest to a fresh change address, which pays another small amount and sends the rest on again 1. Once the change rule is trusted, an analyst can walk the chain hop by hop and read off every recipient. Following the first 100 hops of three peel chains that started from one large wallet, Meiklejohn’s team found that 54 of the 300 peels went to exchanges 1. Their conclusion was that an agency with subpoena power would be well placed to learn who pays whom 1.

Commercial tools now do this at scale, and courts have examined them. Ruling on the Bitcoin Fog prosecution in February 2024, Judge Randolph Moss found that the government had shown its blockchain analysis was “the product of reliable principles and methods” and said such analysis “is not junk science” 2. The software had attributed more than 900,000 addresses to the mixer. The same opinion notes that the case did not rest “exclusively, or even primarily” on it: seized devices, forum posts, IP records and manual tracing carried as much weight 2. That is the realistic picture. Chain analysis produces leads and corroboration. It rarely stands alone.

What a mixer breaks, and what it leaves

A mixer tries to cut the line between coins going in and coins coming out. There are three designs, and each has a different place where it can be taken apart.

A custodial mixer takes your coins into a pool and later pays out different coins. The link exists only in the operator’s records, which makes the operator the weak point. When ChipMixer was taken down on 15 March 2023, police seized four servers, about 7 TB of data and about 1,909 BTC, then worth about EUR 44 million. Europol said the service, running since mid-2017, may have laundered 152,000 BTC 3. The seized data is the record the mixer existed to hide. The pattern is older than ChipMixer. When the Dutch fiscal police took Bestmixer.io offline on 22 May 2019, they seized six servers in the Netherlands and Luxembourg, said the service had handled at least $200 million in a year, and announced that they had gathered information on transactions between customers and the mixer, to be analysed with Europol and then shared with other countries 5. Keeping no records is not a way out either. In October 2020 FinCEN fined Larry Dean Harmon, operator of the Helix mixer, $60 million. Among its findings: he never collected names or addresses on more than 1.2 million transactions and “actively deleted even the minimal customer information” he did collect 6.

The operators themselves are usually named by other people’s records. Minh Quoc Nguyen, charged in March 2023 as ChipMixer’s operator, had repeatedly told forum readers “please do not use AML/KYC exchanges”, and once wrote of Coinbase handing data to the IRS: “That’s what KYC exchanges lead to. All your data will be shipped to IRS.” The complaint against him is built from the business records of hosting companies, PayPal and Google, covering accounts that ran the service under false names. A PayPal account that paid monthly for one of ChipMixer’s servers received a $150 card payment from “MINH NGUYEN”, and records from Binance showed an account in that name, verified with a Vietnamese government ID card 7. Bitcoin Fog ran from 2011 to 2021 and processed more than 1.2 million BTC, about $400 million at the time. Investigators obtained records of its operator’s true-name accounts at several exchanges and traced most of what was deposited there back to Bitcoin Fog’s clusters 8. Roman Sterlingov was convicted by a jury in March 2024 and sentenced on 8 November 2024 to 12 years and 6 months 9. He maintained that he was only a user 10.

mixer · volume claimedactionthe mixer's own recordsthe operatorBestmixer.io$200M+ in one year22 May 2019six servers seized (NL, LU)customer transactions collected,to be shared with other countriesnot named in theannouncementHelix$311M+, 2014–201719 Oct 2020FinCEN: $60M penaltycustomer data deleted;FinCEN cited the deletionpenalty on the operator,Larry Dean HarmonBitcoin Fog1.2M+ BTC, 2011–202127 Apr 2021operator arrestedwallet of about 1,345 BTClater forfeitedhis true-name exchange accounts,funded from Fog's clustersTornado Cash$7B+ since 20198 Aug 2022sanctioned (lifted 2025)immutable contracts:no operator log to seizedeveloper convicted onone count, 2025ChipMixer153,732 BTC in, 2017–202315 Mar 2023takedown, Germany and USback-end servers and7 TB of data seizeda $150 card payment in his name;his ID card at Binance
Five mixer cases, one row each. Where an operator held the deposit-to-withdrawal link, seizing its servers made that link readable to investigators (yellow); a smart contract left no such log (black). In two cases the operator was named by records outside the mixer: exchange accounts, a card payment, an ID card (magenta). Volumes are the authorities' figures.

CoinJoin removes the custodian. Many users sign one shared transaction with equal-sized outputs, which defeats the common-input rule because the inputs really do belong to different people. Someone still has to organise the rounds, though. Samourai Wallet ran such a coordinator. In November 2025 its founders were sentenced, Keonne Rodriguez to five years and William Lonergan Hill to four, for a service that prosecutors said handled more than $237 million in illegal transactions 11. The coordinator is the part a court can reach.

A smart-contract mixer has no operator holding funds at all. Users deposit into a contract and later withdraw to a new address with a zero-knowledge proof. Tornado Cash is the example. Treasury had sanctioned its first mixer, Blender.io, on 6 May 2022, for processing over $20.5 million from the Axie Infinity theft 12. On 8 August 2022 it sanctioned Tornado Cash, saying it had laundered more than $7 billion since 2019, including over $455 million stolen by the North Korean Lazarus Group 13. On 26 November 2024 the Fifth Circuit held that its immutable smart contracts were not “property” that sanctions could block 14, and on 21 March 2025 Treasury removed the sanctions 15. With no server to seize, the criminal case went to a person. In August 2025 a jury convicted developer Roman Storm on one count, conspiracy to operate an unlicensed money-transmitting business, and could not agree on the laundering and sanctions counts. As reported in August 2026, the retrial on those counts is set for 26 April 2027 while his motion for acquittal is still pending 16.

mixer poola custodial operator logs this pathin 1.00 BTC02:58in 0.50 BTC03:02in 2.37 BTC03:10out 0.99 BTC03:31out 0.49 BTC04:15out 2.35 BTC03:52similar size, minutes apart: linked anywayExchangeaccount holderhidden: which deposit became which withdrawalreadable: amounts, times, the exitnames someone: the exchange's record
What a mixer hides and what it leaves. The path inside the pool is hidden from the public, but amounts, times and the exit are not, and a custodial operator keeps the hidden path in its own records.

Whatever the design, a mixer only hides the path inside it. The amounts going in and out, the times, and above all where the coins go next stay on the public ledger. A deposit of an unusual size followed shortly by a withdrawal of nearly the same size is a lead no matter how many other users were in the pool.

Privacy coins: sound math, leaky use

Privacy coins try to fix the ledger itself. Monero hides the real input of each transaction among decoys, called mixins, and gives every payment a one-time address; the blockchain basics page shows the mechanics. Two studies in 2017 measured how well this worked on the chain as it then was, and both found the weakness in how it was used rather than in the cryptography.

The key fact is that an output can be spent only once. Kumar, Fischer, Tople and Saxena analysed every transaction up to 6 February 2017 and found that 65.9% of all inputs had no decoys at all, which made them “traceable by default”. Each of those reveals an output that has been spent, so wherever that output appears as a decoy in another ring it can be crossed out. Repeating the step brought the total to 87.9% of inputs traced 17. Möser and colleagues, with data to 15 April 2017, found the same cascade: 64.04% of inputs had no decoys, and about 62% of the inputs that did have one or more could be deduced. They also estimated that simply guessing the newest ring member would pick the real input about 80% of the time across all inputs with decoys 18.

Chain reactionring 1out ano decoys: a is the real inputring 2out aout ba was spent in ring 1, so b is realring 3out bout cb is spent too, so c is realWhat the studies measured, and whenminimum ring size (real input + decoys)RingCTdecoys optional≥ 3≥ 5≥ 7111620142015201620172018201920202021202220232024Kumar et al. · data to Feb 201788% of inputs traced with certaintyMöser et al. · data to Apr 201762% of inputs with decoys deducednewest ring member was the real one ~80% of the timethe measured data ends before these rules
Left: the chain reaction. A ring with no decoys gives away its real input, and that spent output can then be crossed out wherever it appears as a decoy. Right: the data both studies used ends in early 2017, before Monero made RingCT mandatory and raised the minimum ring size. Yellow is the share an analyst could trace, black the share still hidden.

The era matters. Neither cascade touched the newer RingCT transactions: Kumar’s attack “did not find results on RingCTs” 17, and Möser’s team put that down not to RingCT itself but to the fact that it arrived after a two-decoy minimum was already in force 18. The rules then tightened. From 16 September 2017 only RingCT transactions with rings of at least five were allowed, then at least seven from April 2018, exactly 11 from October 2018 and 16 from August 2022 19. The September 2017 release also changed how decoys are picked, based on an early draft of the Möser paper 18. The well-known percentages are a measurement of early Monero, not of Monero today.

Zcash offers a shielded pool alongside ordinary transparent addresses, and the problem there was how few people used it. Kappos, Yousaf, Maller and Meiklejohn parsed the chain up to 21 January 2018. 73.5% of transactions were fully transparent, and only 14.96% touched the shielded pool at all. Linking deposits by the founders and by miners to later withdrawals accounted for 65.6% of the value withdrawn from the pool, and their simple heuristics shrank the overall anonymity set by 69.1% 20.

Even a coin that holds up inside the chain is visible at its edges. In the Incognito Market case, the market’s administrator wallet sent bitcoin to a swap service four times between July 2021 and May 2022. Each time, 21 to 40 minutes later, an exchange account received Monero of about the same value: 2 BTC went in on 31 May 2022, were swapped for about 304.74 XMR, and 300 XMR arrived in the account 35 minutes later. The account had been opened with a Taiwanese driver’s licence 21. Nobody had to trace anything inside Monero. The way in and the way out were enough.

The other limit is practical. A coin that is hard to trace is also hard to sell at a regulated venue. Binance, for one, delisted Monero in February 2024 22. That pushes holders back toward a conversion step, and conversion is where the next section begins.

The exit: where the ledger meets a name

Coins on a chain do not pay rent. At some point they have to become ordinary money, and the businesses that do that conversion are where the ledger meets an identity document. A regulated exchange asks for ID when an account is opened. Clustering tells an investigator that one owner sent coins to that exchange; the exchange’s records say who the owner is.

The volume through that door is large. Chainalysis estimates that flows from illicit sources to centralized exchanges have averaged more than $14 billion a year since 2020. What has changed is the route: direct transfers from illicit addresses made up over 40% of those deposits in 2021 and 2022 and around 15% by the second quarter of 2025, so more money now passes through intermediaries before it arrives 23. The intermediaries are increasingly professional. Chainalysis counted $16.1 billion through Chinese-language laundering networks in 2025, about 20% of the laundering it could see, and it calls its overall total a lower bound because it only covers services the company has attributed 24.

Services that skip those checks are the obvious alternative, and they become targets in their own right. US prosecutors say the Russian exchange Garantex processed at least $96 billion in cryptocurrency transactions from April 2019. On 6 March 2025 the US Secret Service seized its domains, German and Finnish police took its servers, and more than $26 million was frozen 25. A look-alike called Grinex appeared within weeks, and Treasury designated it as Garantex’s successor in August 2025 26. Rebrands are cheap, but the old and new wallets are connected on the ledger, which is how analysts tied the two together.

Stablecoins add a different kind of chokepoint. A token like USDT is issued by a company that can freeze balances at specific addresses. Just before the Garantex takedown, Tether froze about $27 million of USDT held in its wallets, and the exchange halted withdrawals 27. A chain can be permissionless while the asset on it is not.

Where it gives way

The documented cases fail at the edges of the chain, not in the middle.

case · on the public ledgermeant to break the trailwho knew the customerwhat their records heldWelcome to Video2015–2018420+ BTC in 7,300+ paymentsa new deposit address for everyuser (1.3 million+ addresses)Exchanges 1–3where users paid fromaccount holders, matched tousernames on the seized serverTwitter hackJuly 2020the “Chaewon” wallet clusterhandles on a forum and Discord;no names on the chainBinance, Coinbaseaccounts tied to the walletsdriver's licence,date of birth, addressBitfinex2016–2022119,754 BTC stolenpeel chains, AlphaBay pass-through,fake-name accounts, chain hoppingsix other exchangesaccounts in their own nameslogins from the sameNew York IP addressesIncognito Market2021–2022the admin walletBTC swapped to Monero, thendeposited 21–40 minutes laterone exchange account(“Crypto Account-1”)Taiwanese driver's licence,phone number, email
Four cases drawn the same way. What was on the public ledger (yellow) went through steps meant to break the trail, and every row still ends at a business that knew its customer, whose records named someone (magenta).
  • Welcome to Video. The site took at least 420 BTC through about 7,300 transactions. IRS-CI followed the payments to the operator and the server in South Korea, and 337 users were charged around the world 28. The site gave every user a fresh deposit address, more than 1.3 million in all. The forfeiture complaint shows how that was undone: the payments came from accounts at three exchanges, and the customer data on the seized server showed which username had made which payment 29.
  • Twitter, 2020. Agents traced about 3.69 BTC from the “Chaewon” wallet to another member of the group. Binance records showed that accounts linked to that wallet were controlled by Mason Sheppard, and Coinbase’s know-your-customer data “included Mason Sheppard’s driver’s license, date of birth, and address” 34.
  • Bitfinex. 119,754 BTC were stolen in 2016. In February 2022 agents seized about 94,000 BTC that were still sitting in the wallet that had received them, after getting into files that held the private keys 30. The coins had been visible to everyone for six years. The part that had moved had already run into exchanges. Accounts opened under other people’s names were frozen in 2017 when the exchange asked for more identification, and a spreadsheet in Ilya Lichtenstein’s cloud storage listed them as “FROZEN”. Funds also reached accounts in the couple’s own names at six other exchanges, and a $500 Walmart gift card bought with traced coins was ordered from an IP address leased in Lichtenstein’s name 31.
  • Incognito Market. Besides the Monero deposits described above, the administrator wallet helped pay for a domain bought through a Namecheap account registered to “RuiSiang Lin”, with a Taiwanese phone number and a Taipei address 21.
  • Colonial Pipeline. Of the roughly 75 BTC paid in May 2021, the Justice Department seized 63.7 BTC on 7 June after following transfers on the public ledger to an address for which the FBI had the private key. It did not say how it got the key 32.
  • Bybit. On 21 February 2025 about $1.5 billion in crypto was stolen. Five days later the FBI attributed it to North Korea’s “TraderTraitor” actors, published laundering addresses, and noted that the thieves were “proceeding rapidly”, dispersing converted assets “across thousands of addresses on multiple blockchains” 33. Within about ten days all of the stolen ether had been swapped, mostly through a cross-chain protocol. Bybit’s chief executive said about 77% of the funds were still traceable and about 3% had been frozen 34. Tracing worked. Recovery needs someone at the other end who can freeze, and the thieves moved faster than that.

The pattern is the same throughout. The cryptography held. What gave way was a key, a custodian, or the step where coins had to turn into something spendable.

Common misconceptions

  • “Bitcoin is anonymous.” It is pseudonymous. Every payment is public, and clustering has been documented in the open literature since 2013 1.
  • “A mixer makes coins untraceable.” It hides one path. Operators and coordinators get seized or prosecuted, their servers hand over the path they hid, and amounts, timing and the exit stay public 4 5 11.
  • “Tornado Cash was delisted, so the case is over.” Removing sanctions is not an acquittal. Roman Storm stands convicted on one count, with a retrial on the others scheduled 16.
  • “Studies showed Monero is traceable.” They showed that early Monero was, through missing decoys and predictable decoy choice, on data that ends in early 2017; the rules changed after that 17 18. Zcash’s weakness was that most people never used its private pool 20.
  • “Traceable means recoverable.” Bybit was traced in near real time and most of it still got away 33.

Cases that failed here

Sources

Show all 34 sources
  1. ACM Internet Measurement Conference (Meiklejohn et al., UC San Diego)A Fistful of Bitcoins: Characterizing Payments Among Men with No Namesprimary
  2. U.S. District Court for the District of Columbia (via FindLaw)United States v. Sterlingov, No. 21-399 (RDM), memorandum opinion on blockchain-analysis evidenceprimary
  3. EuropolOne of the darkweb's largest cryptocurrency laundromats washed outprimary
  4. BleepingComputerChipMixer platform seized for laundering ransomware payments, drug sales
  5. FIOD (Fiscal Information and Investigation Service, Netherlands)The FIOD and the Public Prosecution Service take money laundering machine for cryptocurrencies offlineprimary
  6. Financial Crimes Enforcement Network (FinCEN)First Bitcoin "Mixer" Penalized by FinCEN for Violating Anti-Money Laundering Lawsprimary
  7. U.S. Department of Justice (FBI affidavit in support of criminal complaint)United States v. Minh Quoc Nguyen, No. 2:23-mj-00528, criminal complaint and affidavitprimary
  8. U.S. Attorney's Office, District of ColumbiaJury Finds Russian-Swedish Operator of Bitcoin Fog Guilty of Running Darknet Cryptocurrency Mixerprimary
  9. IRS Criminal InvestigationBitcoin Fog operator sentenced for money laundering conspiracyprimary
  10. Infosecurity MagazineMan Gets 12.5 Years for Running Bitcoin Fog Crypto Mixer
  11. IRS Criminal InvestigationFounders of Samourai Wallet cryptocurrency mixing service sentenced to five and four years in prisonprimary
  12. U.S. Department of the TreasuryU.S. Treasury Issues First-Ever Sanctions on a Virtual Currency Mixer, Targets DPRK Cyber Threatsprimary
  13. U.S. Department of the TreasuryU.S. Treasury Sanctions Notorious Virtual Currency Mixer Tornado Cashprimary
  14. U.S. Court of Appeals for the Fifth Circuit (via Justia)Van Loon v. Department of the Treasury, No. 23-50669primary
  15. U.S. Department of the TreasuryTornado Cash Delistingprimary
  16. The BlockTornado Cash developer Roman Storm's retrial pushed back to April 2027
  17. ESORICS 2017 (Kumar, Fischer, Tople and Saxena, National University of Singapore)A Traceability Analysis of Monero's Blockchainprimary
  18. Proceedings on Privacy Enhancing Technologies (Möser et al.)An Empirical Analysis of Traceability in the Monero Blockchainprimary
  19. The Monero Project (GitHub)Monero README: scheduled software and network upgrades (hard fork table)primary
  20. USENIX Security Symposium (Kappos, Yousaf, Maller, Meiklejohn)An Empirical Analysis of Anonymity in Zcashprimary
  21. U.S. District Court for the Southern District of New York (via U.S. Department of Justice)United States v. Rui-Siang Lin, 24 Mag. 351, sealed complaintprimary
  22. The BlockBinance to delist Monero after previously adding 'monitoring tag'
  23. ChainalysisThe Landscape of Seizable Crypto Assets in 2025
  24. ChainalysisThe Chinese-language Underground Crypto Money Laundering Ecosystem (2026 Crypto Crime Report)
  25. U.S. Department of JusticeGarantex Cryptocurrency Exchange Disrupted in International Operationprimary
  26. ChainalysisLegislated Sanctions Evasion: How the Garantex Rebrand, Grinex, and the Ruble-Backed Token, A7A5 Have Shaped Russia's Shadow Crypto Economy
  27. CointelegraphTether freezes $27M USDT on sanctioned Russian exchange Garantex
  28. U.S. Department of JusticeSouth Korean National and Hundreds of Others Charged Worldwide in the Takedown of the Largest Darknet Child Pornography Websiteprimary
  29. U.S. District Court for the District of Columbia (via U.S. Department of Justice)United States v. Twenty-Four Cryptocurrency Accounts, No. 1:19-cv-03098, verified complaint for forfeiture in remprimary
  30. U.S. Department of JusticeTwo Arrested for Alleged Conspiracy to Launder $4.5 Billion in Stolen Cryptocurrencyprimary
  31. U.S. District Court for the District of Columbia (via U.S. Department of Justice)United States v. Lichtenstein and Morgan, No. 1:22-mj-00022, statement of factsprimary
  32. U.S. Department of JusticeDepartment of Justice Seizes $2.3 Million in Cryptocurrency Paid to the Ransomware Extortionists Darksideprimary
  33. FBI Internet Crime Complaint Center (IC3)North Korea Responsible for $1.5 Billion Bybit Hackprimary
  34. CointelegraphBybit hacker launders 100% of stolen $1.4B crypto in 10 days