Pith. sign in

REVIEW 4 major objections 6 minor 73 references

AMLGuard claims that crypto money laundering on EVM blockchains can be traced continuously and precisely by first lifting raw token transfers into high-level DeFi intents, and reports 94.4% destination precision and 98.4% address recall on

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

AMLGuard recovers high-level DeFi intent from raw token flows, including cross-chain intent, and uses it to trace laundered crypto with 94.4%/87.6% destination precision and 98.4%/95.8% address recall on single-chain/cross-chain cases.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection A genuinely useful semantic-lifting AML tracer whose reported numbers depend on author-built ground truth sharing the method's assumptions; deserves peer review but needs independent labels and open artifacts. the 4 major comments →

arxiv 2607.18869 v1 pith:3TDLBFJ6 submitted 2026-07-21 cs.CR

Tracing the Shadows: Automatic Tracking and Analysis of Crypto Money Laundering via Transaction Semantic Analysis

Hao Wu , Haijun Wang , Shangwang Li , Yin Wu , Ming Fan , Ting Liu , Xiapu Luo This is my paper
classification cs.CR
keywords anti-money launderingtransaction semantic analysisDeFi Semantic Unittoken flow graphcross-chain tracingLLM reasoningEVM account-based blockchainslaundering topology
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper asks whether illicit cryptocurrency flows can be followed automatically from a known hacker address all the way to exit services like exchanges, despite the noise of DeFi transactions and the fragmentation of cross-chain bridges. It answers yes with AMLGuard: instead of chasing every token transfer, the system abstracts each transaction into a DeFi Semantic Unit — a compact representation of who initiated which financial operation, with which tokens, and where the trace should go next. Rule-based graph search handles common operations (swap, deposit, withdraw, transfer, approve); a retrieval-augmented LLM with consistency checks handles the rest; and cross-chain intent is recovered by parsing calldata arguments to find the destination chain and recipient address. On 82 real-world incidents, AMLGuard reconstructs compact laundering topologies with destination precision of 94.4% (single-chain) and 87.6% (cross-chain), address recall of 98.4% and 95.8%, and cuts manual investigation time by over 60% in a user study. A sympathetic reader would care because if this holds, automated AML can move from noisy graph heuristics to semantics-driven tracing that an analyst can actually verify.

Core claim

The central claim is that transaction semantics are the right unit of AML tracing: a single external transaction that spawns over a hundred internal token movements can be distilled into one financial intent, and this distillation is both sufficient and necessary for accurate tracing. The paper's core discovery is that five DeFi operation types — approve, transfer, swap, deposit, withdraw — expressed as a uniform DeFi Semantic Unit (initiator, target, token-in, token-out) capture the laundering-relevant behavior on EVM chains; once a transaction is lifted to this level, protocol-internal noise (routers, fee accounts, mint/burn addresses) can be safely pruned. For cross-chain flows, the paper

What carries the argument

The DeFi Semantic Unit (DSU) — an abstraction <Operation, Initiator, Target, InToken, OutToken> that records a transaction's financial intent while discarding intermediate execution steps — is the load-bearing concept. It is produced from a Token Flow Graph (TFG) built from all transfer actions in a transaction, classified either by deterministic graph-search rules matching five canonical operation patterns or by retrieval-augmented LLM reasoning with factual and logical consistency checks that reject implausible inferences. Cross-chain continuity is provided by a second mechanism: proxy-resolution, function-signature matching, and argument semantic parsing that decode bridge calldata into (

Load-bearing premise

The ground-truth laundering paths used to score the system were themselves reconstructed by analysts using the same kind of semantic recovery — decompiling contracts and reading cross-chain bridge documentation — that AMLGuard performs, so the evaluation may reward the system for reproducing the annotators' reasoning rather than finding independently verifiable truths.

What would settle it

A re-evaluation dataset where each laundering destination is verified by off-chain means (exchange KYC records, subpoena-confirmed withdrawals, or court filings) rather than by on-chain semantic reconstruction; if AMLGuard's destination recall on that set falls materially below 94%, the paper's central claim would be weakened — similarly, a bridge protocol that encrypts or obscures recipient arguments in calldata would falsify the cross-chain recovery mechanism's generality.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If AMLGuard's numbers hold, a single known hack address can seed an automated trace that finds roughly 98% of laundering addresses on the same chain and 96% across chains, while keeping the candidate topology small enough for manual review.
  • Cross-chain laundering paths are not semantically opaque: destination chain and recipient can be recovered from bridge calldata with a 94.6% recovery rate on tested incidents, letting a trace continue onto a second ledger without an explicit on-chain link.
  • Distilling transactions into DSUs lets the tracer prune protocol-internal addresses, reducing traced entities from tens of thousands (heuristic baselines) to tens per incident, which is the difference between analysis and noise.
  • The workflow is modular: 22.9% of laundering transactions require LLM inference beyond static rules, and with retrieval augmentation and iterative feedback, operation inference accuracy reaches 96–97% even on protocols held out of the knowledge base.
  • Manual AML investigation time drops by over 60% when auditors work with the system's reconstructed graphs instead of raw explorers, suggesting operational value for freezing funds before cash-out.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the method's discriminative power comes from classifying token-flow patterns, it should extend to any EVM-compatible chain without retraining, since the same five operation types underlie most DeFi protocols; a testable extension would be applying the DSU abstraction to non-EVM chains such as Solana by building a token-flow graph from its instruction logs.
  • The calldata-parsing insight implies a race: if this cross-chain recovery becomes widely known, privacy-preserving bridges or bridges that encrypt recipient parameters would defeat this specific recovery arm; a testable prediction is that such protocol changes will shift laundering toward bridges that do not encode plaintext recipient semantics.
  • The 60% time reduction in the user study points to a complementary direction: using DSUs as a structured query layer for backward tracing from exits (exchanges, mixers) to find unknown seed addresses; the paper flags backward propagation as future work, and the DSU's account-state updates seem directly reusable for that.
  • The evaluation's reliance on semantically recovered ground truth is the key vulnerability: should the dataset's laundering paths turn out to have been built on the same inference the tool performs, the reported precision and recall would overstate field performance — but a controlled re-annotation using only on-chain confirmed exchange deposits would settle this.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper presents AMLGuard, a semantic-aware anti-money laundering framework for EVM-compatible blockchains. It lifts raw transaction data into Token Flow Graphs (TFGs), infers high-level DeFi operation semantics using a combination of static rules and retrieval-augmented LLM reasoning, and abstracts each transaction into a DeFi Semantic Unit (DSU). Cross-chain semantic recovery is performed via function signature matching and argument parsing of calldata, enabling laundering paths to be traced across ledgers. Starting from a known malicious address, AMLGuard iteratively expands the tracing frontier in a depth-first manner to reconstruct compact illicit fund-flow topologies. The evaluation is on a newly constructed dataset of 82 real-world laundering incidents (63 single-chain, 19 cross-chain). AMLGuard reports destination precision of 94.4% (single-chain) and 87.6% (cross-chain), address recall of 98.4% and 95.8%, and destination recall of 94.1% and 93.8%, outperforming five baseline methods. Ablation studies quantify the contribution of the semantic analysis pipeline and the LLM-based inference module, and a user study suggests practical benefit in reducing manual investigation time.

Significance. If the reported results hold, AMLGuard would be a substantial advance in crypto AML: it addresses the semantic gap in DeFi transaction analysis, enables cross-chain tracing, and produces compact, investigator-friendly topologies. The paper's strengths include a real-world evaluation dataset spanning seven chains and over $1B in illicit assets, a clear component-wise ablation, a separate generalization test for LLM-based inference on unseen protocols, and an open-source replication package. The cross-chain semantic recovery module, in particular, addresses a genuinely open problem. However, the validity of the headline metrics depends critically on the independence and correctness of the ground-truth laundering paths, and that independence is not established.

major comments (4)
  1. [Section 5.1 (Dataset) and Table 2] The ground-truth laundering paths were constructed by the authors using blockchain explorers, decompilation, and cross-chain protocol documentation to 'semantically recover interrupted cross-chain transactions.' This is the same kind of semantic inference that AMLGuard performs. The reported Kappa=0.932 only demonstrates inter-annotator agreement, not that the labels are independent of the method's assumptions. MetaSleuth is itself an AML tool and, as the paper notes, 'may miss certain flows,' so it cannot validate completeness or precision. Since all recall/precision numbers in Table 2 depend on this ground truth, the evaluation may be circular. Please provide an independent validation set (e.g., paths confirmed by security firms' post-hoc investigations, or labels produced by annotators who are blind to the method and do not use semantic-recovery reasoning), and report metrics on that
  2. [Table 2 and Section 5.2 (RQ1)] Destination precision is the headline metric for filtering quality, yet it is not reported for four of the five baselines (TRacer, Haircut, Poison, APPR). The text says direct precision measurement is 'impractical' because these methods produce large, noisy graphs. This makes the claim 'destination precision of 94.4% and 87.6% ... outperforming existing methods' incomplete: the comparison on the filtering dimension is missing for most baselines. Please report precision for the baselines using a consistent criterion (e.g., top-k destinations, or manual inspection of a subsample of destinations), or explicitly state that precision is only compared against XBlockFlow and state the resulting limitation.
  3. [Section 5.2, last paragraph (false-negative analysis)] The paper states that to 'mitigate path explosion,' AMLGuard excludes addresses whose outgoing transfer amount exceeds their incoming transfer amount 'based on a heuristic algorithm.' This exclusion rule directly affects both precision and recall: it may suppress false positives that would otherwise be counted, and it is also cited as a source of false negatives. The rule is an ad-hoc parameter of the system, not a reported design choice with sensitivity analysis. Please specify the heuristic precisely, justify it independently of the evaluation set, and report how results vary as the exclusion threshold is varied. Without this, the compactness and precision of AMLGuard's topologies are not fully attributable to semantic analysis.
  4. [Section 4.2.2 (five core DeFi operations)] The paper asserts that five DeFi operations (Approve, Transfer, Swap, Deposit, Withdraw) 'capture all laundering-relevant semantics.' This is a load-bearing assumption: any laundering behavior outside these categories is either misclassified or skipped. The assertion is based on a review of DeFiLlama top categories, but no quantitative coverage analysis is provided (e.g., what fraction of the 82 incidents' ground-truth operations fall into these five classes, and are there any observed operations that were re-labeled or dropped?). Since the ground-truth annotation process may use similar semantic categories, this assumption could also influence the evaluation. Please report the distribution of operation types in the dataset and any instances that required extension of the taxonomy.
minor comments (6)
  1. [Global] Typographical errors: 'strcutured' (Section 4.4.2), 'normlized' (Section 4.4.2), 'Damonds' in reference [24], 'Januray' in multiple references, and 'EIP-2535 diamond' formatting. Please run a spell-check pass.
  2. [Algorithm 1] Line 10 pushes 'DSU.𝑟' onto the stack, but Definition 4.3 uses 'R' for the target entity; the notation is inconsistent. Also, the algorithm's termination condition and the handling of addresses that appear multiple times are not fully explained.
  3. [Figure 2] Figure 2 contains duplicated blocks (e.g., 'Iterative Loop' appears twice, and some labels such as 'Operation Pattern Database Matching' appear in both the rule-based and LLM-based branches). Please clean up the figure to avoid confusion.
  4. [Section 5.5 (Case Study)] The sentence 'In both settings, experts are allowed to use blockchain explorers' is inconsistent with the three settings listed. The user study reports time reductions but does not report the accuracy of the auditors' final conclusions; adding an accuracy measure (e.g., fraction of ground-truth destinations found) would strengthen the claim.
  5. [Section 5.4.3 (Table 4)] The table reports 'Tx Succ' but the text says 'recovery accuracy of 94.6%' (424/448). Please clarify whether 'Succ' means fully recovered semantics or merely identified destination, and define the failure cases more precisely.
  6. [References] Some references are incomplete (e.g., [5] 'Web3 API' with no author, [41] 'The Prototype of AMLGuard'). The arXiv identifier in the header is 2607.18869, which is inconsistent with the '2026' publication year; please verify the metadata.

Circularity Check

1 steps flagged

No derivation-level circularity; the only mild circularity is evaluative: the cross-chain ground-truth paths were reconstructed with the same semantic-recovery approach AMLGuard implements, so reported cross-chain metrics partly measure agreement with the annotators' own recovery method.

specific steps
  1. other [Section 5.1 (Experiment Setup, Dataset construction) and Section 4.4 (Cross-Chain Semantic Recovery)]
    "During this process, we further leveraged the decompilation provided by blockchain explorers and cross-chain protocol documentation to semantically recover interrupted cross-chain transactions."

    The cross-chain evaluation (Table 4 reports 94.6% recovery accuracy; Table 2 reports Dc destination recall of 93.8%) is scored against ground-truth paths that the annotators themselves produced by semantically recovering interrupted cross-chain transactions—the same kind of inference AMLGuard's cross-chain semantic recovery performs. Thus the high cross-chain agreement partly reflects the annotators applying the method's own semantic-recovery priors, not an independently constructed ground truth. This is a structural evaluation concern, not a mathematically forced fit: AMLGuard still must perform the inference, so the metrics are not determined by construction. It is a mild evaluative circularity rather than a derivation-level one.

full rationale

AMLGuard is an empirical system evaluated on a manually curated dataset; there is no derivation chain in which a prediction is defined as a fitted input or in which a claimed result reduces to an input by construction. The paper's central numbers—address recall, destination recall, destination precision—are measurements against annotated laundering traces, not derived consequences of the method's definitions. Self-citations (e.g., the authors' prior smart-contract analysis works) are not load-bearing for the AML tracing claim. The RAG-based LLM inference is additionally tested on previously unseen protocols with independently annotated labels, providing some evidence of generalization beyond the knowledge base. The one genuine circularity concern is evaluative: Section 5.1 states that ground-truth cross-chain paths were 'semantically recovered' by the annotators using decompilation and protocol documentation, which mirrors AMLGuard's cross-chain semantic recovery. Section 6 candidly acknowledges that 'Our ground-truth dataset is constructed based on manual analysis, which may introduce potential inaccuracies' and that annotation accuracy 'directly impacts the effectiveness of AMLGuard.' That limitation is real but does not reduce the paper's derivation to its inputs; it mainly weakens the independence of the cross-chain benchmark. Accordingly, the circularity score is low (2), reflecting a mild evaluation-circularity concern rather than a forced or definitional circularity.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

AMLGuard introduces the DSU abstraction and a five-operation taxonomy as internal software constructs, not physical entities requiring external evidence. The central claim depends on hand-chosen parameters (retrieval count, iteration limit, tracing heuristics) and on domain assumptions about the sufficiency of single-transaction analysis, the completeness of the operation taxonomy, and the reliability of blockchain explorers and protocol documentation.

free parameters (4)
  • RAG retrieval top-k = 2
    Number of similar TFG examples fed to the LLM; chosen by hand in Section 5.1 and affects semantic inference accuracy.
  • LLM refinement iteration limit = 3
    Maximum correction loops in Section 4.3.2; chosen by hand, affects accuracy and cost.
  • Intermediary exclusion rule = outgoing amount > incoming amount
    Heuristic in Section 5.2 that excludes addresses from tracing to mitigate path explosion; directly affects false positives and recall.
  • Ground-truth path value threshold = unspecified
    Section 5.1 excludes 'transfer paths with negligible value' from ground truth, which can bias precision metrics.
axioms (4)
  • domain assumption Only a few DeFi operations span multiple transactions; single-transaction analysis suffices
    Section 4.2 states this based on study [72], narrowing scope to single-tx granularity.
  • ad hoc to paper Five core DeFi operations (Approve, Transfer, Swap, Deposit, Withdraw) capture all laundering-relevant semantics
    Section 4.2.2 defines this taxonomy from DeFiLlama top 15 types; completeness is assumed for AML coverage.
  • domain assumption EVM event logs and explorer APIs provide complete and correct token transfer information
    Section 4.2.1 and 5.1 rely on these data sources without independent verification.
  • domain assumption Cross-chain function signature templates from Chainspot are complete and accurate
    Section 4.4 relies on this curated set for signature matching; missing templates cause failures.

reviewed 2026-08-01 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Tracing the Shadows: Automatic Tracking and Analysis of Crypto Money Laundering via Transaction Semantic Analysis." pith.science (2026). https://pith.science/paper/3TDLBFJ6

@misc{pith2026260718869,
  author       = {Pith},
  title        = {Pith review of: Tracing the Shadows: Automatic Tracking and Analysis of Crypto Money Laundering via Transaction Semantic Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3TDLBFJ6}},
  note         = {Machine review of arXiv:2607.18869}
}
Share X Bluesky LinkedIn Reddit HN
abstract

With the rapid advancement of decentralized finance (DeFi), security incidents related to cryptocurrency have become increasingly prevalent. After such incidents, attackers typically attempt to rapidly move stolen assets, concealing the origin of illicit funds and ultimately converting them into fiat currency. However, existing anti-money laundering (AML) methods struggle to cope with the semantic complexity of DeFi transactions. They either rely heavily on low-level token transfers, or perform protocol-agnostic money flow analysis, failing to capture the high-level intent of transactions. In this paper, we propose AMLGuard, a semantic-aware AML framework for account-based blockchains. AMLGuard tracks illicit fund flows from known malicious addresses by performing semantic analysis on complex DeFi transactions, enabling accurate and continuous laundering tracking. Given a complex transaction, AMLGuard combines static rule-based analysis with retrieval-augmented large language model (LLM) reasoning to infer implicit DeFi semantics, transforming raw transaction data into high-level semantic representations. Furthermore, for cross-chain transactions where laundering intent is not explicitly exposed, AMLGuard parses transaction parameters and performs argument parsing to recover cross-chain semantics, enabling seamless tracking across ledgers. Based on the inferred semantics, AMLGuard abstracts each transaction into a DeFi Semantic Unit (DSU). We evaluate the effectiveness of AMLGuard on 82 real-world laundering cases, involving illicit assets worth over $1 billion. Specifically, AMLGuard reconstructs compact illicit fund-flow topologies with destination precision of 94.4% and 87.6%, while achieving the highest address recall of 98.4% and 95.8% and destination recall of 94.1% and 93.8% on single-chain and cross-chain datasets.

Figures

Figures reproduced from arXiv: 2607.18869 by Haijun Wang, Hao Wu, Ming Fan, Shangwang Li, Ting Liu, Xiapu Luo, Yin Wu.

Figure 1
Figure 1. Figure 1: Money Laundering Process of Cream Finance Incident [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of AMLGuard We introduce AMLGuard, an iterative framework to track crypto money laundering through transaction semantic analysis. As shown in [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: An Example Showing the Workflow of Transaction Semantic Analysis [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: LLM-based DeFi Semantic Inference 4.3.1 Retrieval-Augmented Operation Infernce. We adopt a RAG pipeline to ground LLM-based semantic inference in domain knowledge. The core idea is to expose the LLM to previously analyzed DeFi operations with similar token-flow structures, enabling analogy-driven reasoning. To this end, we construct a DeFi Operation Knowledge Base by serializing TFGs and computing their em… view at source ↗
Figure 5
Figure 5. Figure 5: An Example Showing the Workflow of Cross-Chain Semantic Recovery [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Efficiency Comparison on Datasets 𝐷𝑠 and 𝐷𝑐 Compared to baseline methods, AMLGuard performs fine-grained transaction semantic analysis, which introduces additional time overhead. To quantify this overhead, we further evaluate the average analysis time per transaction (TPT). As shown in [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Tracing Visualization for Li.Fi Incident [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

73 extracted references · 2 linked inside Pith

  1. [1]

    Ismail Alarab, Simant Prakoonwit, and Mohamed Ikbal Nacer. 2020. Competence of graph convolutional networks for anti-money laundering in bitcoin blockchain. InProceedings of the 2020 5th international conference on machine learning technologies. 23–27. , Vol. 1, No. 1, Article . Publication date: July 2026. AMLGuard: Automatic Tracking and Analysis of Cry...

    2020

  2. [2]

    Reem Alshammari, Rwan Alrashed, Atheer Almutiri, Mathail Alwalah, Wadha Al-marrai, Dana Alqahtani, and Amani Alzahrani. 2021. Data extraction based on web scrapy. InInnovations in Bio-Inspired Computing and Applications: Proceedings of the 11th International Conference on Innovations in Bio-Inspired Computing and Applications (IBICA 2020) held during Dece...

    2021

  3. [3]

    Erik Altman, Jovan Blanuša, Luc Von Niederhäusern, Béni Egressy, Andreea Anghel, and Kubilay Atasu. 2023. Realistic synthetic financial transactions for anti-money laundering models.Advances in Neural Information Processing Systems 36 (2023), 29851–29874

    2023

  4. [4]

    Reid Andersen, Fan Chung, and Kevin Lang. 2006. Local graph partitioning using pagerank vectors. In2006 47th annual IEEE symposium on foundations of computer science (FOCS’06). IEEE, 475–486

    2006

  5. [5]

    Web3 API. 2026. https://web3py.readthedocs.io/en/stable/web3.main.html. Accessed: January, 2026

    2026

  6. [6]

    Simon Joseph Aquilina, Fran Casino, Mark Vella, Joshua Ellul, and Constantinos Patsakis. 2021. EtherClue: Digital investigation of attacks on Ethereum smart contracts.Blockchain: Research and Applications2, 4 (2021), 100028

    2021

  7. [7]

    Arbitrum. 2026. Arbitrum. https://arbitrum.io/. Accessed: January, 2026

    2026

  8. [8]

    Beosin. 2024. Li.Fi Incident. https://beosin.com/resources/beosin-trace-analyzes-the-10-million-loss-of-lifi-protocol. Accessed: Januray, 2026

    2024

  9. [9]

    Beosin. 2026. https://beosin.com/. Accessed: January, 2026

    2026

  10. [10]

    BlockSec. 2024. MetaSleuth. https://metasleuth.io/. Accessed: January, 2026

    2024

  11. [11]

    BlockSec. 2026. https://blocksec.com/. Accessed: January, 2026

    2026

  12. [12]

    BSC. 2026. https://www.bnbchain.org/. Accessed: January, 2026

    2026

  13. [13]

    Vitalik Buterin, Jacob Illum, Matthias Nadler, Fabian Schär, and Ameen Soleimani. 2024. Blockchain privacy and regulatory compliance: Towards a practical equilibrium.Blockchain: Research and Applications5, 1 (2024), 100176

    2024

  14. [14]

    placement, layering, integration

    Stefan D Cassella. 2018. Toward a new model of money laundering: Is the “placement, layering, integration” model obsolete?Journal of Money Laundering Control21, 4 (2018), 494–497

    2018

  15. [15]

    Certik. 2024. Cream Finance Incident Hacker Address. https://etherscan.io/address/ 0x49B9eb77B300014F99b39b35904c2DBc069e428E. Accessed: January, 2026

    2024

  16. [16]

    Certik. 2025. 0xInfini Incident. https://www.certik.com/resources/blog/0xinfini-incident-analysis. Accessed: January, 2026

    2025

  17. [17]

    2025.Hack3d: The Web3 Security Report 2025

    Certik. 2025.Hack3d: The Web3 Security Report 2025. Accessed: May, 2025. https://www.certik.com/resources/blog/ hack3d-the-web3-security-report-2025

    2025

  18. [18]

    Certik. 2026. https://www.certik.com/. Accessed: Januray, 2026

    2026

  19. [19]

    Certik. 2026. Cream Finance Attack Incident. https://x.com/CreamdotFinance/status/1453455806075006976. Accessed: Januray, 2026

    arXiv 2026

  20. [20]

    Chainspot. 2025. https://chainspot.io/portal/bridges. Accessed: Januray, 2026

    2025

  21. [21]

    Ravindu De Silva, Wenbo Guo, Nicola Ruaro, Ilya Grishchenko, Christopher Kruegel, and Giovanni Vigna. 2024. {GuideEnricher}: Protecting the Anonymity of Ethereum Mixing Service Users with Deep Reinforcement Learning. In 33rd USENIX Security Symposium (USENIX Security 24). 3549–3566

    2024

  22. [22]

    DefiLlama. 2025. https://defillama.com/. Accessed: Januray, 2026

    2025

  23. [23]

    Hanbiao Du, Zheng Che, Meng Shen, Liehuang Zhu, and Jiankun Hu. 2023. Breaking the anonymity of ethereum mixing services using graph feature learning.IEEE Transactions on Information Forensics and Security19 (2023), 616–631

    2023

  24. [24]

    Multi-Facet Proxy EIP-2535: Damonds. 2025. https://eips.ethereum.org/EIPS/eip-2535. Accessed: Januray, 2026

    2025

  25. [25]

    Ethereum. 2026. https://ethereum.org/en/. Accessed: January, 2026

    2026

  26. [26]

    Etherscan. 2026. https://etherscan.io/. Accessed: January, 2026

    2026

  27. [27]

    Steven Farrugia, Joshua Ellul, and George Azzopardi. 2020. Detection of illicit accounts over the Ethereum blockchain. Expert Systems with Applications150 (2020), 113318

    2020

  28. [28]

    Josselin Feist, Gustavo Grieco, and Alex Groce. 2019. Slither: a static analysis framework for smart contracts. In2019 IEEE/ACM 2nd International Workshop on Emerging Trends in Software Engineering for Blockchain (WETSEB). IEEE, 8–15

    2019

  29. [29]

    Chris Gilbert and Mercy Gilbert. 2024. Unlocking privacy in blockchain: Exploring zero-knowledge proofs and secure multi-party computation techniques.A vailable at SSRN 5258791(2024)

    2024

  30. [30]

    Yicheng Huo, Yufeng Hu, Yajin Zhou, Ting Yu, Lei Wu, and Cong Wang. 2025. Shedding light on shadows: Automatically tracing illicit money flows on EVM-compatible blockchains.Proceedings of the ACM on Measurement and Analysis of Computing Systems9, 3 (2025), 1–35

    2025

  31. [31]

    Ziqiao Kong, Cen Zhang, Maoyi Xie, Ming Hu, Yue Xue, Ye Liu, Haijun Wang, and Yang Liu. 2025. Smart contract fuzzing towards profitable vulnerabilities.Proceedings of the ACM on Software Engineering2, FSE (2025), 153–175

    2025

  32. [32]

    Dan Lin, Jiajing Wu, Yuxin Su, Ziye Zheng, Yuhong Nan, Qinnan Zhang, Bowen Song, and Zibin Zheng. 2025. Connector: Enhancing the traceability of decentralized bridge applications via automatic cross-chain transaction association.IEEE , Vol. 1, No. 1, Article . Publication date: July 2026. 22 H. Wu, H. Wang, S. Li, Y. Wu, M. Fan, T. Liu, and X. Luo Transac...

    2025

  33. [33]

    Dan Lin, Jiajing Wu, Yunmei Yu, Qishuang Fu, Zibin Zheng, and Changlin Yang. 2024. DenseFlow: Spotting cryptocur- rency money laundering in ethereum transaction graphs. InProceedings of the ACM Web Conference 2024. 4429–4438

    2024

  34. [34]

    Dan Lin, Ziye Zheng, Jiajing Wu, Jingjing Yang, Kaixin Lin, Huan Xiao, Bowen Song, and Zibin Zheng. 2025. Track and trace: Automatically uncovering cross-chain transactions in the multi-blockchain ecosystems.IEEE Transactions on Services Computing(2025)

    2025

  35. [35]

    Han Liu, Daoyuan Wu, Yuqiang Sun, Haijun Wang, Kaixuan Li, Yang Liu, and Yixiang Chen. 2024. Using My Functions Should Follow My Checks: Understanding and Detecting Insecure{OpenZeppelin} Code in Smart Contracts. In33rd USENIX Security Symposium (USENIX Security 24). 3585–3601

    2024

  36. [36]

    Jiayi Liu, Changchun Yin, Hao Wang, Xiaofei Wu, Dongwan Lan, Lu Zhou, and Chunpeng Ge. 2023. Graph embedding- based money laundering detection for Ethereum.Electronics12, 14 (2023), 3180

    2023

  37. [37]

    Ye Liu, Yi Li, Shang-Wei Lin, and Cyrille Artho. 2022. Finding permission bugs in smart contracts with role mining. In Proceedings of the 31st ACM SIGSOFT International Symposium on Software Testing and Analysis. 716–727

    2022

  38. [38]

    Joana Lorenz, Maria Inês Silva, David Aparício, João Tiago Ascensão, and Pedro Bizarro. 2020. Machine learning methods to detect money laundering in the bitcoin blockchain in the presence of label scarcity. InProceedings of the first ACM international conference on AI in finance. 1–8

    2020

  39. [39]

    Bruno Mazorra, Victor Adan, and Vanesa Daza. 2022. Do not rug on me: Leveraging machine learning techniques for automated scam detection.Mathematics10, 6 (2022), 949

    2022

  40. [40]

    Malte Möser, Rainer Böhme, and Dominic Breuker. 2014. Towards risk scoring of Bitcoin transactions. InFinancial Cryptography and Data Security: FC 2014 Workshops, BITCOIN and W AHC 2014, Christ Church, Barbados, May 7, 2014, Revised Selected Papers 18. Springer, 16–32

    2014

  41. [41]

    The Prototype of AMLGuard. 2026. https://figshare.com/s/e01b691346bb352701e5. Accessed: Januray, 2026

    2026

  42. [42]

    OpenAI. 2026. https://openai.com/. Accessed: Januray, 2026

    2026

  43. [43]

    Bofeng Pan, Natalia Stakhanova, and Zhongwen Zhu. 2024. EtherShield: Time-interval Analysis for Detection of Malicious Behavior on Ethereum.ACM Transactions on Internet Technology21, 1 (2024), 1–30

    2024

  44. [44]

    SharkTeam. 2026. https://sharkteam.org/. Accessed: January, 2026

    2026

  45. [45]

    EIP-1967: Proxy Storage Slots. 2025. https://eips.ethereum.org/EIPS/eip-1967. Accessed: Januray, 2026

    1967

  46. [46]

    SlowMist. 2026. https://slowmist.com/. Accessed: January, 2026

    2026

  47. [47]

    Jie Song, Sijia Zhang, Pengyi Zhang, Junghoon Park, Yu Gu, and Ge Yu. 2024. Illicit Social Accounts? Anti-Money Laundering for Transactional Blockchains.IEEE Transactions on Information Forensics and Security(2024)

    2024

  48. [48]

    Jianzhong Su, Xingwei Lin, Zhiyuan Fang, Zhirong Zhu, Jiachi Chen, Zibin Zheng, Wei Lv, and Jiashui Wang. 2023. Defiwarder: Protecting defi apps from token leaking vulnerabilities. In2023 38th IEEE/ACM International Conference on Automated Software Engineering (ASE). IEEE, 1664–1675

    2023

  49. [49]

    Yuqiang Sun, Daoyuan Wu, Yue Xue, Han Liu, Haijun Wang, Zhengzi Xu, Xiaofei Xie, and Yang Liu. 2024. Gptscan: Detecting logic vulnerabilities in smart contracts by combining gpt with program analysis. InProceedings of the IEEE/ACM 46th international conference on software engineering. 1–13

    2024

  50. [50]

    Redeem Transaction. 2022. https://etherscan.io/tx/0x8d714c2fdcd90eb44b075d34233ead8b97fe6fcf13b2428520b58fd57feba52f. Accessed: January, 2026

    2022

  51. [51]

    Stake Transaction. 2022. https://etherscan.io/tx/0x34eec380808c47d699587eaa904bacb71bfc35d9a0de6cfae6a19b8bac217af6. Accessed: January, 2026

    2022

  52. [52]

    Dylan Vassallo, Vincent Vella, and Joshua Ellul. 2021. Application of gradient boosting algorithms for anti-money laundering in cryptocurrencies.SN Computer Science2, 3 (2021), 143

    2021

  53. [53]

    Haijun Wang, Yurui Hu, Hao Wu, Dijun Liu, Chenyang Peng, Yin Wu, Ming Fan, and Ting Liu. 2024. Skyeye: Detecting imminent attacks via analyzing adversarial smart contracts. InProceedings of the 39th IEEE/ACM International Conference on Automated Software Engineering. 1570–1582

    2024

  54. [54]

    Haijun Wang, Ye Liu, Yi Li, Shang-Wei Lin, Cyrille Artho, Lei Ma, and Yang Liu. 2020. Oracle-supported dynamic exploit generation for smart contracts.IEEE Transactions on Dependable and Secure Computing19, 3 (2020), 1795–1809

    2020

  55. [55]

    Zhipeng Wang, Stefanos Chaliasos, Kaihua Qin, Liyi Zhou, Lifeng Gao, Pascal Berrang, Benjamin Livshits, and Arthur Gervais. 2023. On how zero-knowledge proof blockchain mixers improve, and worsen user privacy. InProceedings of the ACM Web Conference 2023. 2022–2032

    2023

  56. [56]

    Mark Weber, Giacomo Domeniconi, Jie Chen, Daniel Karl I Weidele, Claudio Bellei, Tom Robinson, and Charles E Leiserson. 2019. Anti-money laundering in bitcoin: Experimenting with graph convolutional networks for financial forensics.arXiv preprint arXiv:1908.02591

    Pith/arXiv arXiv 2019

  57. [57]

    Cong Wu, Jing Chen, Ziming Zhao, Kun He, Guowen Xu, Yueming Wu, Haijun Wang, Hongwei Li, Yang Liu, and Yang Xiang. 2024. Tokenscout: Early detection of ethereum scam tokens via temporal graph learning. InProceedings of the 2024 on ACM SIGSAC Conference on Computer and Communications Security. 956–970. , Vol. 1, No. 1, Article . Publication date: July 2026...

    2024

  58. [58]

    Hao Wu, Haijun Wang, Shangwang Li, Yin Wu, Ming Fan, Wuxia Jin, and Ting Liu. 2025. RPHunter: Unveiling Rug Pull Schemes in Crypto Token via Code-and-Transaction Fusion Analysis.arXiv preprint arXiv:2506.18398(2025)

    Pith/arXiv arXiv 2025

  59. [59]

    Hao Wu, Haijun Wang, Shangwang Li, Yin Wu, Ming Fan, Yitao Zhao, and Ting Liu. 2025. Detecting State Manipulation Vulnerabilities in Smart Contracts Using LLM and Static Analysis. InProceedings of the 16th International Conference on Internetware. 317–320

    2025

  60. [60]

    Jiajing Wu, Dan Lin, Qishuang Fu, Shuo Yang, Ting Chen, Zibin Zheng, and Bowen Song. 2023. Toward understanding asset flows in crypto money laundering through the lenses of Ethereum heists.IEEE Transactions on Information Forensics and Security19 (2023), 1994–2009

    2023

  61. [61]

    Jiajing Wu, Kaixin Lin, Dan Lin, Bozhao Zhang, Zhiying Wu, and Jianzhong Su. 2025. Safeguarding blockchain ecosystem: Understanding and detecting attack transactions on cross-chain bridges. InProceedings of the ACM on Web Conference 2025. 4902–4912

    2025

  62. [62]

    Lei Wu, Yufeng Hu, Yajin Zhou, Haoyu Wang, Xiapu Luo, Zhi Wang, Fan Zhang, and Kui Ren. 2021. Towards understanding and demystifying bitcoin mixing services. InProceedings of the Web Conference 2021. 33–44

    2021

  63. [63]

    Siwei Wu, Zhou Yu, Dabao Wang, Yajin Zhou, Lei Wu, Haoyu Wang, and Xingliang Yuan. 2024. DeFiRanger: Detecting DeFi Price Manipulation Attacks.IEEE Transactions on Dependable and Secure Computing21, 4 (2024), 4147–4161

    2024

  64. [64]

    Yin Wu, Xiaofei Xie, Chenyang Peng, Dijun Liu, Hao Wu, Ming Fan, Ting Liu, and Haijun Wang. 2024. Advscanner: Generating adversarial smart contracts to exploit reentrancy vulnerabilities using llm and static analysis. InProceedings of the 39th IEEE/ACM International Conference on Automated Software Engineering. 1019–1031

    2024

  65. [65]

    Zhiying Wu, Jieli Liu, Jiajing Wu, Zibin Zheng, and Ting Chen. 2023. TRacer: Scalable graph-based transaction tracing for account-based blockchain trading systems.IEEE Transactions on Information Forensics and Security18 (2023), 2609–2621

    2023

  66. [66]

    Zhiying Wu, Jieli Liu, Jiajing Wu, Zibin Zheng, Xiapu Luo, and Ting Chen. 2023. Know your transactions: Real-time and generic transaction semantic representation on blockchain & web3 ecosystem. InProceedings of the ACM Web Conference 2023. 1918–1927

    2023

  67. [67]

    Maoyi Xie, Ming Hu, Ziqiao Kong, Cen Zhang, Yebo Feng, Haijun Wang, Yue Xue, Hao Zhang, Ye Liu, and Yang Liu

  68. [68]

    Haaroon Yousaf, George Kappos, and Sarah Meiklejohn. 2019. Tracing transactions across cryptocurrency ledgers. In 28th USENIX Security Symposium (USENIX Security 19). 837–850

    2019

  69. [69]

    Mengya Zhang, Xiaokuan Zhang, Yinqian Zhang, and Zhiqiang Lin. 2020. {TXSPECTOR}: Uncovering attacks in ethereum from transactions. In29th USENIX Security Symposium (USENIX Security 20). 2775–2792

    2020

  70. [70]

    Zhuo Zhang, Zhiqiang Lin, Marcelo Morales, Xiangyu Zhang, and Kaiyuan Zhang. 2023. Your exploit is mine: Instantly synthesizing counterattack smart contract. In32nd USENIX Security Symposium (USENIX Security 23). 1757–1774

    2023

  71. [71]

    Zongyang Zhang, Jiayuan Yin, Bin Hu, Ting Gao, Weihan Li, Qianhong Wu, and Jianwei Liu. 2022. CLTracer: A Cross-Ledger Tracing framework based on address relationships.Computers & Security113 (2022), 102558

    2022

  72. [72]

    Juantao Zhong, Daoyuan Wu, Ye Liu, Maoyi Xie, Yang Liu, Yi Li, and Ning Liu. 2025. Detecting Various DeFi Price Manipulations with LLM Reasoning. In2025 40th IEEE/ACM International Conference on Automated Software Engineering (ASE). 1781–1793. , Vol. 1, No. 1, Article . Publication date: July 2026

    2025

  73. [2024]

    InProceedings of the 33rd ACM SIGSOFT International Symposium on Software Testing and Analysis

    Defort: Automatic detection and analysis of price manipulation attacks in defi applications. InProceedings of the 33rd ACM SIGSOFT International Symposium on Software Testing and Analysis. 402–414

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.