# FPTracer Finds 98.6% of Labeled Ethereum Laundering Addresses

Published 2026-10-26 · Crypto.world Team · Crypto.world

> A paper submitted to arXiv on October 23 reports that FPTracer recovered 98.6% of labeled laundering addresses across 10 Ethereum cases. The system also identified bridge activity and flagged possible Tornado Cash withdrawal-side flows in a 20-case study.

A research team’s FPTracer system recovered 98.6% of labeled laundering addresses in a 10-case evaluation of Ethereum tracing methods, according to a paper submitted to [arXiv](https://arxiv.org/abs/2610.32210) on October 23. The result measures address recall in cases with labeled laundering traces; it is not a standalone claim that the system detects 98.6% of all cross-chain money laundering.

## What the 98.6% measures

The study evaluates FPTracer on 20 real-world laundering incidents from 2019 to 2025, with reported losses totaling about $5.47 billion. For the address-tracing comparison, the researchers use 10 of those cases that overlap with an existing dataset of annotated laundering traces. FPTracer recovered 98.6% of labeled addresses and reached 94.2% of labeled destinations, the highest recall figures among the methods compared.

> On the 10 cases with labeled laundering traces, FPTracer achieves the highest recall among the compared methods, identifying 98.6% of labeled laundering addresses and 94.2% of labeled laundering destinations.
>
> Huan Xing, Gaohe Jin, Bang Fu, Yue Li, Baoning Niu and Xing Fan, Taiyuan University of Technology

The comparison included XBlockFlow, which recorded 90.2% address recall and 84.7% destination recall, and AMLGuard-Public, with 95.3% and 87%, respectively. The authors define recall as the share of labeled laundering addresses recovered; destination recall measures the share of labeled destinations reached during tracing. Those results describe recovery against the study’s reference labels, not a measure of how often an alert against an unknown address is correct.

## FPTracer follows asset paths inside transactions

FPTracer’s method starts with a problem in Ethereum transaction records: a complex DeFi transaction can generate many transfers across contracts, obscuring how value moves from its source to its actual recipient. The system builds a Context-Aware Transfer Tree, or CATT, that places transfers within the nested contract calls that produced them. It then links transfers into asset-flow paths using execution order, addresses, token types and value continuity.

Those reconstructed paths guide the expansion of a laundering network. A separate classifier examines asset-flow and execution evidence, including event logs and storage changes, to identify transactions with cross-chain intent without relying only on a precompiled list of bridge signatures. For mixers, FPTracer uses evidence from the upstream network to flag possible withdrawal-side activity; the paper presents those outputs as candidates, not deterministic links between deposits and withdrawals.

## Bridge findings carry a ground-truth caveat

For its transaction-level bridge classifier, the team assembled a benchmark of 8,533 cross-chain deposit transactions from 28 bridge protocols and 8,441 non-deposit transactions. In tests that held out bridge protocols from training, the authors report that all classifiers except a decision tree achieved F1 scores above 97%, while results varied more depending on which bridges were held out. XGBoost, selected for the case-level analysis, recorded 97.23% recall in that bridge-disjoint evaluation.

Across the 20 laundering incidents, FPTracer identified all bridge-use pairs documented in the reports the researchers collected, plus 30 additional pairs. The team says it manually checked those additional pairs by tracing assets to labeled laundering addresses and examining bridge explorers, contracts and transaction evidence. The authors also caution that public reports do not provide complete ground truth for bridge use, so the reported coverage is bounded by the documentation available to them.

## Tornado Cash candidates remain investigative leads

The paper says FPTracer identified Tornado Cash deposits in 18 of the 20 cases and generated possible withdrawal-side candidates in seven. Because the mixer breaks the observable on-chain link between deposits and withdrawals, the authors could not establish a definitive correspondence from the chain alone. Public incident analyses provided corroboration for 222 of 442 candidates in the KucoinHacker case and 47 of 82 in the LiquidHacker case; the paper does not classify the other candidates as false positives.

That distinction frames the study’s practical claim: FPTracer can reconstruct more of the traceable flow through DeFi transactions, identify bridge deposits, and surface potential activity after mixer use for investigators to examine. Its headline recall result comes from 10 labeled laundering cases, while bridge and mixer findings rely in part on manual review and incomplete public reporting.

## Sources

1. Original source · Preprint [Tracing Money Laundering Across DeFi, Bridges, and Mixers on Ethereum](https://arxiv.org/abs/2610.32210) arXiv:2610.32210

## Related stories

- [Ethereum Draft Seeks Reproducible Transaction Simulations](https://crypto.world/news/ethereum-draft-reproducible-transaction-simulation-results)
- [Ethereum Draft Defines Risk Reports for AI-Signed Transactions](https://crypto.world/news/ethereum-draft-agent-pre-sign-risk-report)
- [Ethereum Draft Adds CALLSTATUS to Distinguish Reverts From Halts](https://crypto.world/news/ethereum-callstatus-instruction-draft)
- [ENS Launches ENSv2 on Ethereum Mainnet](https://crypto.world/news/ens-launches-ensv2-ethereum-mainnet)
- [Ethereum’s 300-Byte ENR Limit Complicates Post-Quantum Networking](https://crypto.world/news/ethereum-consensus-layer-post-quantum-networking-gap)

## Explore

[More articles](https://crypto.world/news)

---

Canonical: https://crypto.world/news/fptracer-ethereum-laundering-tracing-98-6-recall
