Tornado Cash in 2026: at least 18% of the ETH that went in was stolen, over 40% of that by one attacker — and almost none of it can be traced out

Tornado Cash has four ETH pools: 0.1, 1, 10 and 100 ETH. Every deposit and every withdrawal is a public event on Ethereum. We read all of them from 1 January to 21 October 2026 (06:41 UTC): 64,860 deposits worth 645,194 ETH from 10,912 addresses, and 61,826 withdrawals worth 681,871 ETH. More came out than went in because the pools also pay out deposits made before 2026; on 21 October the 100 ETH pool alone held about 252,700 ETH.

Then we checked every depositor against our database of addresses from published thefts, and for the 100 ETH pool, where 74% of the money goes, we followed every unlabelled depositor up to three steps back, to the first exchange, bridge, DEX or known theft.

87,429 ETH, 13.6% of all deposits, came straight from wallets of published thefts. We count only addresses tied to an incident that has been reported and that we or others traced; one large deposit (3,916 ETH on 23 July) that rests only on a victim’s report to us is left out.

The money is concentrated. The Drift Protocol exploiter put in 48,805 ETH, more than half of the total, in two waves: 23,095 ETH on 23 July and 25,710 ETH on 20 October. Then come Triple-A (7,672 ETH), the Ledger/CryptoBilis drain (5,144), Bonzo Lend (4,571), Summer.fi (3,294), Term Labs (3,253), Tectonic (3,259), the THORChain vault exploit (2,500), Aztec Connect (2,464) and Duelbits (1,370). Most of these we traced ourselves; the list of pages is below.

Following unlabelled depositors of the 100 ETH pool back one to three steps adds about 29,600 ETH that reached them from labelled thefts (among them the Lick and US-seizure thefts, Truebit and address-poisoning scams; these labels come from our database and were not re-checked one by one). We removed a label that marks an MEV bot as an “exploiter”.

Together: at least about 117,000 ETH, roughly 18% of all ETH deposited in 2026, was stolen. This is a floor. The other three pools were not followed back, and about 80,000 ETH in the 100 ETH pool could not be traced to any known source within three steps.

For the 100 ETH pool we can say where most of the unlabelled money entered. By value, January–July and August–October: bridges 17% and 22% (Across, Relay, Chainflip, Stargate, Mayan, NEAR Intents, THORChain), DEX swaps 26% and 17% (mostly CoW Protocol, sellers swapping DAI, USDC and WBTC), other services 8% and 12%, exchanges only 4% and 3%, other Tornado pools 3%. Behind the bridges we could resolve the chain of origin for 65% of the value: Solana, Arbitrum and Base, often people moving their own money between chains.

Two large unlabelled flows stand out. In January, 686 BTC sent through THORChain became 19,631 ETH, and 11,200 ETH of it went into the 100 ETH pool through 23 fresh wallets between 28 January and 20 April; the bitcoin had been gathered from 50-BTC lots that sat untouched since September 2024. In July, 168,000 SOL were swapped through Mayan into 5,116 ETH and deposited. Neither matches a published theft. A three-day burst of 30 deposits in February that looked like layering turned out to come from Japanese exchange withdrawals.

A Tornado withdrawal does not say which deposit it spends; that is the point of the zero-knowledge proof. What leaks is behaviour around the pool. We tested the standard heuristics on all 2026 events: the same address on both sides, a withdrawal paid or funded by the depositor, money that comes back together after the pool, unusual combinations of amounts, timing, gas-price fingerprints.

Only 4.7% of withdrawals (4.4% of the ETH) link to a deposit with confidence, almost all of them repeat users and services moving their own money. Amount and timing patterns add 0.2%. Not one withdrawal links to the Drift, THORChain or Aztec deposits: the attackers used relayers and fresh addresses and never reused one.

If we accept guesses from timing alone, almost every withdrawal gets a candidate: for about 47% of the ETH withdrawn, one depositor made more than half of the deposits into that pool in the previous 24 hours. But the pool holds around 250,000 ETH of older deposits at any time, so that model overstates the odds by a wide margin, and there is no ground truth to measure it. What we can act on is the next step after a withdrawal.

After the 20 October Drift wave, ten fresh wallets took exactly 100 ETH each within seven minutes, an hour after the deposits ended; each was submitted by its own gas wallet funded through MAYAChain the day before. 500 ETH went on to NEAR Intents in 50 ETH pieces, and 439.9 ETH collected on one wallet, part of which went straight back into Tornado on 21 October.

After the Aztec deposits on 19 October, three wallets took 11 × 100 ETH within 15 hours, swapped it to USDT and bridged 2.64 million USDT to TRON through the USDT0 bridge. There the money was mixed with other funds, split into one-million chunks and, in part, swapped USDT → USDD → USDT, which takes it out of reach of a Tether freeze, before reaching deposit addresses of large unnamed payment services.

These are matches by time and amount, not proof. We give such wallets a high risk score as possible recipients of stolen funds, so that compliance teams check them; we do not call their owners thieves. 58 addresses are marked so far.

Volume. TRM Labs wrote on 12 June 2026 that Tornado Cash had received over $700 million in 2026; we count 281,793 ETH deposited by 12 June, which fits that order of magnitude at this year’s prices. TRM puts Tornado at just over 20% of all mixer activity in 2026. L2Beat, via The Defiant, called 23 July the busiest Tornado day of 2026 with 968 deposits; we see 955 deposits into the ETH pools that day, 29,573 ETH, of which the Drift exploiter made 245 deposits and 23,095 ETH, 78%. On 20 October its 25,710 ETH was 83% of the day.

Share of stolen money. Chainalysis found in 2023 that stolen funds were about 8% of everything sent to mixers. Researchers at the St. Louis Fed estimated, for Tornado before the 2022 sanctions, that about 70% of the volume was not linked to crime (as cited by the Cambridge Centre for Alternative Finance; we did not read the primary). Our floor of 18% for 2026 sits between these, and it is driven by a few very large thefts: without Drift, the direct share would be 6.0%.

Who deposits. The Cambridge Centre for Alternative Finance (February 2026) found that the share of Tornado depositors funded straight from exchanges fell from about 14% to under 3% after the sanctions. We see the same: exchanges are 3–4% of the money entering the 100 ETH pool; bridges and DEX swaps are now the main on-ramps.

Linking. A 2025 paper on Tornado reported linking 5.1–12.6% of withdrawals with address reuse and transaction links, and 15–22 points more with first-in-first-out timing; it was later withdrawn by its authors over reference errors. A June 2026 study of Railgun links 17.65% of withdrawals uniquely using five heuristics. Our 4.7% is lower because we count only links we can show on chain and leave timing as a guess.

All theft wallets in this study are in our database at high risk. Since 20 October every address that receives money directly from a Tornado pool also gets a high score (80). The Drift wallet that still holds 56,568 ETH, and the wallets above, are on our watchlist; we will update the case pages when the money moves.

Our investigations of the thefts named here: Drift Protocol, Triple-A, Ledger/CryptoBilis, Bonzo Lend, Summer.fi, Term Labs, Tectonic, Aztec Connect, Duelbits, Payy, Nomic, and the earlier Tornado Cash Q3 2026 count.

To check an address yourself, use the address check.

It does not say that Tornado Cash users are criminals: most of the money in 2026 has no link to a known theft. It does not name any person, and it does not claim any withdrawal as an attacker’s. The figures from TRM Labs, L2Beat/The Defiant, Chainalysis, the St. Louis Fed, Cambridge and the papers are theirs; the deposit counts, the theft shares, the origin breakdown and the linkage rates are ours, read on 21–22 October 2026.