FPTracer recalls 98.6% of laundering addresses and uncovers 30 new bridge uses
The claim
FPTracer can accurately trace illicit assets across Ethereum services with high recall
Published as: “Tracing Money Laundering Across DeFi, Bridges, and Mixers on Ethereum”
So what
The framework enables more complete tracing of illicit crypto flows on Ethereum, aiding investigators and regulators in uncovering hidden laundering steps.
The five-minute version
FPTracer recovers intra-transaction asset-flow paths with 99.33% exact recovery and achieves 98.6% recall of labeled laundering addresses and 94.2% destination recall across 20 real-world cases, also uncovering all reported bridge uses plus 30 undocumented ones and detecting mixer activity. The evidence comes from a 300-transaction manual annotation and 20 case studies, showing strong but not perfect performance, with two failures on uncommon protocol designs and reliance on manual verification for bridge/mixer analysis.
What they found 5
- Exact intra-transaction path recovery rate of 99.33%
“FPTracer exactly recovers 298 of 300 transactions, yielding a Transaction Exact Recovery Rate (TER) of 99.33%.”
- Recall of labeled laundering addresses is 98.6% and destination recall is 94.2%
“FPTracer achieves a Recall of 98.6% and a D-Recall of 94.2%, indicating effective downstream tracing based on recovered intra-transaction asset-flow paths.”
- Identifies all publicly reported bridge uses and 30 additional undocumented uses
“Across all 20 cases, FPTracer identifies all publicly reported bridge uses and discovers 30 additional uses not documented in public reports.”
- Finds Tornado Cash usage in 18 cases and potential withdrawal-side laundering in 7 cases
“FPTracer further finds Tornado Cash usage in 18 cases and identifies potential withdrawal-side laundering activities in 7 cases.”
- Corroborates 222 of 442 and 47 of 82 candidate withdrawal transactions for two high-profile hacks
“For the KucoinHacker and LiquidHacker cases, we find public reports corroborating 222 of 442 and 47 of 82 candidate withdrawal transactions, respectively.”
What undercuts it the paper’s own results
- Two of the 300 sampled transactions were not correctly recovered due to uncommon protocol execution designs.
- Recall and destination recall are below 100%, leaving some labeled laundering addresses and destinations undiscovered.
Stated limits 2
- Both failures arise from uncommon protocol execution design. For example, Sky LitePSM uses different contracts within the same protocol to handle the input and output assets. As a result, the destination of the input transfer differs from the source of the output transfer, violating the shared intermediate address condition required by R2 and causing the recovered path to be fragmented.
- For the case-level bridge analysis in RQ3 and the mixer analysis in RQ4, no comprehensive public benchmarks are available.
What’s new here 4
- Introduces the Context-Aware Transfer Tree (CATT) to reconstruct hidden asset-flow paths within complex Ethereum transactions without contract-specific knowledge.
- Uses a learning-based classifier to identify cross-chain deposit intent from general behavioral patterns, avoiding manual bridge signatures.
- Detects mixer involvement and flags potential post-mixer laundering flows, preserving evidence beyond the mixer exit.
- Discovers previously undocumented bridge usages in real cases, expanding known laundering networks.
Artifacts & methods 9
- FPTracersystem
- Context-Aware Transfer Tree (CATT)primitive
- Cross-chain deposit classifier
- Dcasedataset
- Bpathdataset
- Btracedataset
- Bbridgedataset
- XBlockFlowsystem
- AMLGuard-Publicsystem