MMath thesis defence

Inside the Crypto Laundering Machine

A taxonomic study of crypto money laundering schemes and countermeasures

Hesam Sarkhosh
Master of Mathematics in Computer Science, University of Waterloo
Supervisor: Prof. Diogo Barradas · Committee: Prof. Bernard Wong, Prof. Mohammad Hajiabadi
11 September 2026

The paradox

Mallory has just stolen 1,000 ETH.

Every move makes from now on is public. Forever. To anyone with a browser.

will still get away with it.

This thesis is a map of how.

[1:15] So let's start with Mallory. She has just stolen a thousand ether, say from a hacked protocol. (click) Here is the strange thing about her situation. Every transaction she makes from this moment on trying to move the well earned assets to a safe/hidden place, is written to a public ledger. Anyone with a browser can open a block explorer and follow it. There is no bank secrecy. The trail is there for everyone, forever. (click) And yet, she will most likely get away with it. Billions of dollars is moved each year through exactly this kind of trail and come out the other side looking clean. (click) That is the paradox this thesis is about. A system that is completely transparent, and yet launders money like a charm. The thesis is a map of how this happens, what tools are available for Mallory to achieve her goal and of what stands in her way.

What laundering is

Three stages, in any economy

[1:15] So let's start with the basics. What money laundering actually is? Laundering is a process widely recognized as having three stages Placement, Layering, and Integration also known as PLI stages. (click) Placement is getting dirty money into the financial system. In the classic picture that is cash deposited into a business account. In crypto we see it as on-ramping any form of illicit fiat into any form of crypto. (click) Layering is burying the trail. Shell companies, wire transfers across borders, anything that puts distance between the money and the original source. (click) Integration is the final stage when the layered money comes to rest as some property with a presumably clean story: a house, a painting, a watch. Some form of asset that when someone starts digging into Malloryies life, they either find a plausible story for how she got it, or they find nothing at all.

Why Web3

Why it is a better washing machine

Pseudonymous

Addresses, not names. The ledger shows who paid whom, never who they are.

Permissionless

No account opening, no questions asked. Anyone can use any protocol.

Borderless

One network, two hundred jurisdictions, and rules that do not match.

No custodian

No one holds the funds, so there is no one to subpoena, and nothing to freeze.

Every property that makes DeFi open also makes it easy to abuse.

each property, shown here

No name, no bank, no border, no boss. Where do I sign?

[1:25] So why is Web3 such a lovely washing machine for Mallory? There are four properties in the ecosystem that make the Mallory's life easier. (click) First, a Blockchain is typically pseudonymous; meaning that while the ledger typically shows the trace of the assets publicly, linking those addresses to real identities can become difficult or if done right, impossible. (click) It is permissionless. There is no account to open and no form to fill. And in most cases, no one, not the Canadian Government, nor even the founding fathers of the protocol, can stop Mallory from using their ecosystem. (click) It is borderless. One network runs across two hundred jurisdictions whose rules do not match, and history proves, launderers tend to pick the weakest ones to operate in. (click) And in decentralized finance there is often no custodian at all. Meaning no central entity holds a backdoor to access the funds just in case, so if Mallory wants to move them, there is no one to subpoena and nothing to freeze. (click) Every property that makes DeFi open is also a property that makes it easy to abuse. That is the tension the whole thesis sits on.

The gap

Prior work covers pieces, not the machine

Our contribution: a taxonomy that separates what wants from how gets it.

Method

A systematic review, in five lines

IdentifyGoogle Scholar keyword searches, university library, and grey literature from regulators and industry

50+academic papers and preprints from keyword searches

200+records after snowballing, screened by title, abstract, then full text

2013 to
mid-2025from the first mixer studies to the current state of DeFi

Taxonomybuilt iteratively, PRISMA-informed, checked against prior surveys

Primer, one of two

Four rungs of anonymity

* Confidential transactions hide the amount, not the trail. Some sources treat this as a property of its own, others as a layer that sits on top of full anonymity.

[1:40] Two pieces of vocabulary from the primer, and then we get to the taxonomy. The first is how anonymous a launderer can actually get. On the right is a single transaction record, and each rung takes one more field off it. (click) Mallory starts at the bottom. Pseudonymity. She has an address, not a name. That is the default on Bitcoin and Ethereum, and it is weaker than it sounds, because addresses cluster and eventually touch a service that knows a name. (click) One level higher is set anonymity. She hides in a crowd. Her payment is signed in a way that makes any member of a group an equally plausible sender. (click) The next rung is full anonymity. Now the path itself is hidden. The sender and the route are cryptographically gone. (click) And at the top, confidential transactions hide the amount too, so even the size of the payment is gone. I have put an asterisk on that one, because it is a slightly different kind of property. It covers the amount rather than the trail, so some sources treat it as separate from anonymity, and others assume it sits on top of full anonymity. Every tool later in the talk moves Mallory up some number of rungs, and that is how I will describe them.

Primer, two of two

What the investigator can see

Lens one

On-chain

Everything the ledger records, public and permanent.

addresses and amounts
timestamps and fees
contract events and token moves

Lens two

Off-chain

What custodians and platforms know, reachable by legal process.

identity records from exchanges
IP addresses and device logs
chats, forums, dark web listings

the gap

Lens three

Cross-chain

The traces left when value crosses from one chain to another.

bridge lock, mint, and burn logs
swap contracts and timelocks
often partial, sometimes missing

Three lenses. Nobody indexes them together.

[1:10] The second piece of vocabulary is what an investigator can actually see. The thesis sorts the evidence into three lenses. (click) The on-chain lens is the ledger itself. Addresses, amounts, timestamps, contract events. It is public, permanent, and enormous. (click) The off-chain lens is what the centralized parts of the ecosystem know. An exchange that did identity checks, the IP addresses and device logs it kept, the chat where a deal was arranged. This is reachable, but only through legal process, and only when there is a custodian to ask. (click) The third lens is cross-chain. When value moves from one chain to another, it leaves traces on both sides, in bridge logs or swap contracts, but those traces are often partial and sometimes missing altogether. Nobody indexes them together. Keep this gap in mind. It comes back twice, once in the strategies and once in the infrastructure.

The map

The taxonomy, in two halves

Crypto money laundering schemes

The map

The taxonomy, in two halves

Crypto money laundering schemes

Strategieswhat strategy has

Intermediary mechanismsWhich tools help achieve it

Privacy-enhancing technologiesbuilt to protect privacy

Exploited infrastructurebuilt for something else

Strategies

Mallory's playbook, six moves

1Transaction flow obfuscationMany hops, one trail buried.L

2Commingling's coins, everyone's coins.L

3Disconnected fund flowNo link left on the chain.P L I

4Chain-hoppingNew chain, fresh start.L

5Exploit subjective valuationThe price is whatever we say.L

6Exploit regulatory gapsWhere no sheriff rides.P L I

each move, shown here

Six moves. I never use just one.

[1:50] Here is the left half of the map, the six strategies. Think of them as Mallory's playbook. (click) One, transaction flow obfuscation. She makes the graph noisy. Long chains of hops, peeling small amounts off a large one, dusting many addresses. Nothing is hidden, it is just expensive to follow. (click) Two, commingling. She mixes her coins with everyone else's, so ownership becomes a matter of probability. This is what a mixer does, and it is where the old warehouse banking trick lives on. (click) Three, disconnected fund flow. This is the strong one. Instead of making the link hard to follow, she removes it, so there is nothing on the chain that connects source and destination. It is the only strategy that reaches full anonymity, and it works at every stage, not just layering. I will zoom into it in a moment. (click) Four, chain-hopping. She moves value to another blockchain, and the investigator's cross-chain lens, with its gap, is now the only way to follow. (click) Five, exploiting subjective valuation. Some assets have no reference price. A fifty dollar artwork that sells for fifty thousand is a payment with a story attached. You will see that auction in the infrastructure section. (click) Six, exploiting regulatory gaps. Rules differ by jurisdiction and by platform, and she chooses the weakest one at each step. Together with disconnected fund flow, this is the other strategy that runs through all three stages.

Strategy three, up close

Disjoint reciprocal transfers

Placement

"Got any ETH for sale? I'm paying well. All cash."

"A brand new wallet with one incoming transfer. Nothing suspicious here."

Layering

"I'll buy your BTC with Tether. No middlemen, no fees. Send me your ETH address."

"No link between these wallets. Two chains, and I have to cross-reference everything by hand."

Integration

"Beautiful watch, Alice. If you're selling, I'm paying well, in ETH."

"The wallet looked clean. And the funds went to Alice, who swears she never knew the guy."

A solid arrow is a transfer written to a ledger. A dashed arrow never touches one.

Reciprocal, because each payment pays for the other. Disjoint, because no ledger shows the pair.

[1:40] Let me zoom into strategy three, because it is where the thesis names a scheme that earlier surveys had not written down. We call it disjoint reciprocal transfers. The idea is to split one trade into two separate payments, each of which looks ordinary on its own, and to settle them so that no ledger ever records that the two belong together. (click) At placement, Mallory finds Alice on a forum and buys ether with cash, in person. On the chain, Alice's address simply pays a brand new wallet. That is all an investigator sees. (click) At layering, she does the same trick across two chains. She pays Alice in a stablecoin on Ethereum, and Alice pays her back in Bitcoin, on Bitcoin. Each chain shows one ordinary transfer. The connection between them lives only in a chat message. (click) At integration, she buys a watch from Alice and pays in ether. The wallet has a clean history, and if anyone ever asks, the money went to Alice, who never knew who Mallory was. (click) So the name says what it is. Reciprocal, because each payment pays for the other one. Disjoint, because no ledger shows the pair, and the only record that they belong together is a chat message. That is what disconnected fund flow means, and it is why the cross-chain gap from the primer matters so much.

Privacy-enhancing technologies

Stop saying "mixer"

Zcash's shielded pool and Tornado Cash are the same construction. One is called a privacy coin, the other a mixer.

"Mixer"one word, three different things

CUSTan operator that holds and shuffles deposits
COORDa coordinator that only arranges a joint CoinJoin
APPa smart contract with a shielded pool

"Privacy coin"one word, four places to live

L1a dedicated chain, such as Monero or Zcash
EXTan extension block on a transparent chain
L2a privacy rollup that settles to a base chain
APPa confidential token contract

Axis one: blueprintsthe cryptographic parts a tool is built from

Axis two: deployment formthe one place every tool runs

[1:30] Now the right half of the map, starting with the purpose-built privacy tools. The literature sorts these with two labels, mixers and privacy coins, and I want to argue that both labels hide more than they show. (click) Here is the problem in one sentence. Zcash's shielded pool and Tornado Cash are the same construction. Pooled deposits, and a zero-knowledge proof to withdraw. One of them is called a privacy coin and the other a mixer. (click) Mixer alone is used for three different things. A custodial operator that holds your coins and shuffles them. A coordinator that never touches funds and only arranges a joint transaction. And a smart contract with a shielded pool. Those have completely different legal exposure and completely different failure modes. (click) Privacy coin is no better. The same privacy can live on a dedicated chain, in an extension block, in a rollup, or in a token contract on Ethereum. (click) So the thesis drops the labels and describes every tool on two axes that do not blur. The blueprints it is assembled from, and the single deployment form it runs in. Let me show each axis.

Axis one

Thirteen blueprints, coloured by the strategy they serve

Serves transaction flow obfuscation

TUMBLE

Random delays, fees, and split outputs. Adds noise so timing and amounts stop matching.

Serves commingling

POOLCJ

A shared reserve, or many payments merged into one transaction. Ownership becomes a matter of odds.

Serves disconnected fund flow

RINGSTEALTHZKPSHUFFLEMWBLINDSWAP

The linkage breakers. Hide the sender, hide the receiver, or remove the path between them altogether.

Serves no laundering strategy

CTFHETEE

Confidentiality only. They hide an amount or a contract's state, but who paid whom stays fully visible.

Lego pieces, not a menu. A tool that combines several blueprints inherits the strategies of all of them.

[1:40] Axis one is the blueprints. Thirteen cryptographic or hardware building blocks, and I colour each one by the laundering strategy it serves, using the same colours as the strategy cards. (click) Tumbling, in amber, is the old mixer trick. Random delays, random fees, split outputs. It serves transaction flow obfuscation, because it adds noise without hiding anything. (click) Pooling and CoinJoin, in teal, serve commingling. A shared reserve, or many payments merged into a single transaction, so ownership becomes a matter of probability. (click) The big indigo group serves disconnected fund flow. Ring signatures hide the sender, stealth addresses hide the receiver, zero-knowledge proofs, shuffles, MimbleWimble, blind signatures, and atomic swaps remove the path itself. These are the pieces that move Mallory up the ladder. (click) And a grey group that serves no laundering strategy at all. Confidential transactions, homomorphic encryption, trusted enclaves. They hide an amount or a contract's state, but the graph of who paid whom is untouched. That distinction matters for regulation, and the thesis keeps it explicit. (click) The point is that these are Lego pieces, not a menu. A shielded pool is pooling plus a zero-knowledge proof, so it commingles and disconnects at once. A tool inherits the strategies of every piece it uses.

Axis two

Six deployment forms, every tool in exactly one

CUSTCustodial servicesan operator holds the coins and pays back different ones	Bitcoin Fog POOLTUMBLEHelix POOLTUMBLEChipMixer POOLBlindcoin POOLBLIND
L1Layer-1privacy built into the chain itself	Monero RINGSTEALTHCTZcash POOLZKPGrin MWCTCJNamada POOLZKPCTSTEALTHIncognito RINGSTEALTHCTZKPSecret TEE
EXTChain extensionsan opt-in private ledger beside a transparent one	Litecoin MWEB MWCTCJZerocoin POOLZKP
L2Layer-2 rollupsprivate execution off-chain, proofs on-chain	Aztec POOLZKPZkopru POOLZKPPhenix FHE
APPApplication layeran ordinary smart contract on a transparent chain	Tornado Cash POOLZKPMöbius RINGSTEALTHMixEth SHUFFLEPOOLUmbra STEALTHZama FHE
COORDOff-chain coordinatorusers co-sign one joint transaction, nobody holds funds	Wasabi CJZKPCoinShuffle CJSHUFFLETumbleBit BLINDSWAPObscuro CJTEECoinSwap SWAPStatechains SWAPBLIND

Custodial? No thanks. I have read the court records.

[1:45] Axis two is where the tool runs, and this is the one axis where every tool has exactly one value. Six forms, and I will go quickly, because the names are here for reference, not for reading. The small diagram on the left of each row is where the tool actually runs. Gold means private, grey means transparent. (click) Custodial services. An operator takes your coins, pools them, and pays back different ones. Bitcoin Fog, Helix, ChipMixer. Simple, enormous volumes, and the ones that get seized, because someone is holding the money. (click) Layer-1. Privacy is a property of the chain itself. Monero with ring signatures and stealth addresses, Zcash with a shielded pool, Grin with MimbleWimble, Secret with enclaves. (click) Chain extensions. An opt-in private ledger bolted onto a transparent chain by a soft fork. Litecoin's MWEB is the live example. (click) Layer-2 rollups. Execution moves off-chain and only proofs land on the base chain. Aztec is the canonical one. (click) The application layer. An ordinary smart contract on a transparent chain, reachable with mainstream assets. Tornado Cash lives here, and so do stealth-address and shuffle contracts. (click) And off-chain coordinators. Users co-sign one joint transaction that a coordinator arranges but never funds. Wasabi, CoinShuffle, TumbleBit. Same effect as a custodial mixer, with nobody to arrest. Notice the chips beside each name. Read across a row and you can see, at a glance, which strategies a tool enacts. That is the whole point of the two axes.

One consequence

Custody is what gets seized

I cannot subpoena a smart contract.

Exploited infrastructure

Built for something else. Works for laundering.

Exchanges

12346

P L I

Cross-chain bridges

DeFi mechanisms

1234

P L I

NFTs

135

P L I

Crypto ATMs

P I

Card-based instruments

P L I

Online gambling

126

P L I

ICOs

12346

P L I

Consensus protocols

P L

Payment channel networks

134

1flow obfuscation2commingling3disconnected flow4chain-hopping5subjective valuation6regulatory gapsP L Istages served

None of these were built for . All of them work for .

Spotlight one

The staking loop

Wrap, pledge, borrow, stake. Repeat until clean.

[1:20] First spotlight, decentralized finance, and the pattern I think best shows what is new here. Everything in DeFi is composable, so a launderer can build a loop. Stake ether and receive a liquid staking token. Post that token as collateral in a lending market like Aave. Borrow a stablecoin against it, which looks clean on the surface. Buy more ether with the loan. And then stake that too, which is the first step again, so the loop closes and can run as many times as she likes. Restaking through something like EigenLayer gives her a second tradable token on the way round. (click) After one turn the funds are already behind a swap, a loan, and a token most reviewers do not look at closely. (click) After two turns there are two of each. (click) After three, the illicit origin is still on the chain somewhere, but it is buried under layers of legitimate-looking protocol activity, and nobody in the loop ever held custody. This is commingling, flow obfuscation, and often chain-hopping, all from one strategy of composing protocols that were built for yield. (click)

Spotlight two

Bridges make chain-hopping effortless

[1:05] Second spotlight, bridges. Chain-hopping used to mean going through an exchange, which meant identity checks. A bridge removes that. (click) Mallory locks tokens in a bridge contract on chain A, and an equivalent token is minted to a fresh wallet on chain B. No account, no questions, and the funds are now on a ledger where nobody was watching for them. (click) Remember the third lens from the primer. The evidence is a lock event on one chain and a mint event on another, recorded by different systems in different formats, with nothing that joins them. Investigators have to reconstruct the link by hand, and many analytics tools simply stop at the bridge. That gap is why chain-hopping earns its own strategy, and why bridges are one of the channels earlier surveys largely missed. (click)

Spotlight three

Where the chain meets the bank

Tap to pay. Nobody asks where it came from.

[1:25] Third spotlight. Decentralized finance is mostly a layering machine, because it never touches cash. The crossing points between the chain and the banking world are where placement and integration happen, and two of them are easy to overlook. (click) Crypto ATMs sit in shops and gas stations. Cash goes in, coins come out, often against nothing more than a phone number. That is placement, and TRM Labs puts the illicit share of the cash-to-crypto sector in the hundreds of millions of dollars. (click) Crypto-backed cards run the other way. The issuer converts coins to fiat at the moment of purchase, so the merchant sees an ordinary card payment. That is integration in a single tap. Cards are one of the channels earlier surveys largely missed, and I think they are the most direct route from a hacked wallet to a paid-for dinner. (click) And there is a third route that did not exist a few years ago. The newer non-custodial programs from DeFi neobanks connect a self-custody wallet app to ordinary bank rails like ACH, SEPA, and Pix. Darth pays fiat into the bank and the bank credits Mallory's wallet app, or Mallory sends fiat out of the app through the bank to Darth. Either way the money crosses between the chain and the banking world with no exchange in between, which opens placement and layering through the same account. (click)

Spotlight four

NFTs: the auction room

Art is worth whatever the last bid says.

[1:15] Fourth spotlight, NFTs, and the strategy they carry needs no cryptography at all: exploiting subjective valuation. Darth owes Mallory fifty thousand dollars of illicit money, and they want the payment to look like income. She buys a cheap piece of digital art, a fifty dollar NFT, and puts it up for auction. (click) She bids on it herself from another wallet. Five hundred. (click) Darth, her partner, bids five thousand. Anyone watching sees a bidding war and a hot artist. (click) Fifty thousand. Nobody else in the room has any reason to object, because the room has no idea what this is worth. That is the whole point. There is no reference price for a unique asset. (click) Sold to Darth for fifty thousand and fifty dollars. Darth's illicit payment now arrives as sale proceeds, and Mallory books it as a capital gain on an artwork. On an NFT marketplace this is called wash trading, and it runs between wallets that the same person controls. The art market has worked this way for a long time. NFTs just made it permissionless, global, and settled in seconds. (click)

Countermeasures

Two ways to fight back

My turn.

[1:25] Now the investigators. The thesis sorts countermeasures into two paradigms. (click) Detection finds laundering after it happens. Rule-based systems, thresholds and clustering heuristics, are explainable and match what regulators expect, but they are brittle and drown analysts in false positives. Data-driven methods, mostly graph learning and anomaly detection on the transaction graph, adapt better, but they are starved of reliable labels. And in practice most regulated services run commercial platforms such as Chainalysis, Elliptic, and TRM, so detection has become a compliance industry. (click) Prevention shrinks the opportunity. Regulators, led by the FATF, have extended anti-money-laundering duties to virtual asset service providers. Blocklisting taints coins and addresses. Deanonymization links addresses to people through the off-chain lens and feeds prosecutions. And public awareness cuts off the victims that fake ICOs and in-person trades depend on. (click) Notice that every one of these leans on something a custodian holds, or on a rule a service must follow. That is exactly what non-custodial code and cross-chain movement route around, which brings us to the open problems.

Open challenges

Four problems, one direction each

[1:30] From that synthesis the thesis draws four open challenges, and pairs each with a direction. (click) Data. Detection models are trained on approximate labels, off-chain data sits behind legal process, and cross-chain traces are often missing altogether. The direction is benchmark datasets labelled by regulators and service providers, and models that need fewer labels to begin with. (click) Pseudonymity and decentralization. When there is no custodian and the code cannot be sanctioned, the old levers fail. The direction is privacy tools that are compliance-aware by design, with selective disclosure, and decentralized identity that lets a user prove who they are without a central registry. (click) Regulatory arbitrage. One network, two hundred jurisdictions, and the launderer picks the weakest. The direction is global standards for service providers, and compliance oracles that bring sanctions data into the smart contracts themselves. (click) And legitimate privacy. The same tools that launder money protect dissidents and commercial secrets, so a blanket ban does real harm. The direction is privacy-preserving compliance. The idea is that a user proves a rule was followed, that the funds are not sanctioned, that a limit was respected, without revealing who is behind the address. You prove compliance instead of proving identity. I think that last one is where the most interesting work will happen.

What the map shows

Three things you only see with all of it in view

[1:10] Three things the map shows once you can see all of it at once. (click) Laundering is a chain, never one tool. Mallory composes strategies and mechanisms across all three stages, and countermeasures that target one tool at a time will keep missing the pattern. (click) The centre of gravity has moved. The custodial mixers of the Bitcoin era are being seized and sanctioned, and laundering has become native to Web3, riding on composable protocols, movement across chains, and tokens whose value is easy to overlook. (click) And the ledger cuts both ways. Its transparency is what makes forensics possible at all, and its pseudonymity and privacy tooling are exactly what defeats it. So I do not think the answer is banning privacy. It is accountable privacy, tools that let a user prove compliance without giving up the ledger's other promise.

Contributions

What this thesis gives you

[1:10] To close, the thesis leaves behind one map, drawn in three layers. (click) The first layer is a taxonomy of crypto laundering schemes. Six strategies and the mechanisms that carry them out, mapped to the three stages and to the traces they leave, including schemes that had not been written down before, such as disjoint reciprocal transfers. (click) The second layer lays over the first. A two-axis framework for privacy tools, thirteen blueprints by six deployment forms, which replaces the blurred labels the field has been using. (click) The third layer marks the response. Where detection already works, where prevention stands in the way, and the four places where the map still has holes in it. (click) And it is meant to be used. For investigators, a route from a scheme to the traces it leaves. For people building forensic tools, a view of where coverage is thin. For regulators, a picture of where the gaps actually are.

Thank you

Questions?

Hesam Sarkhosh · Inside the Crypto Laundering Machine · University of Waterloo, 2026

Backup

The full taxonomy

Taxonomy of crypto money laundering schemes

Backup

Transaction flow obfuscation patterns

Backup

Coverage across prior surveys

Backup

Blueprint definitions

TUMBLETumbling. Randomised delays, fees, and output splitting that break timing and amount correlations.

POOLPooling. Funds commingled in a shared reserve or persistent pool; a shielded pool when paired with ZKP or TEE.

CJCoinJoin. Many users' payments merged into one transaction so inputs and outputs cannot be paired.

RINGRing signatures. The true input hidden among decoy signers, so any member is an equally plausible source.

STEALTHStealth addresses. A fresh one-time address per payment, so the receiver's published address never appears on-chain.

ZKPZero-knowledge proofs. Prove ownership of an unspent note without revealing which one.

SHUFFLEShuffle / DC-net. A participant-run permutation, built on multi-party computation, that breaks linkage.

MWMimbleWimble. No addresses; a block's transactions aggregated with cut-through so inputs and outputs cannot be paired.

BLINDBlind signatures. Chaumian tokens that let a coordinator authorise a withdrawal it cannot link to a deposit.

SWAPAtomic swaps. Equal-value coins exchanged through hash time-locked contracts, so no on-chain link is written.

CTConfidential transactions. Commitments and range proofs hide the amount, but not the parties.

FHEFully homomorphic encryption. Computation on encrypted balances and state.

TEETrusted execution environments. Contract state kept inside hardware enclaves that node operators cannot read.

Backup

Review method

Sources

Google Scholar keyword searches: crypto money laundering, AML in Web3, DeFi money laundering, mixing services, and related terms
Librarian-guided search of the finance literature through the university library
Grey literature from regulators (FATF, FinCEN, OFAC, Europol) and industry (Chainalysis, TRM Labs, Elliptic)
Snowballing from references, expanding the pool from 50+ papers to 200+ records

Criteria

Included: topically relevant to Web3-enabled laundering or AML; analytical or technical substance; written in English; 2013 to mid-2025
Excluded: off-topic works such as price prediction; insufficient technical relevance; superseded or duplicate works
Screened by title, then abstract and introduction, then full technical sections
Taxonomy built iteratively and checked against prior surveys