AI in Blockchain Forensics and Crypto Anti-Money Laundering (AML) Compliance

Research focus: AI-powered blockchain analytics, cryptocurrency money laundering detection, transaction monitoring, wallet risk scoring, graph neural networks, suspicious transaction detection, sanctions screening, DeFi AML, cross-chain tracing, stablecoin monitoring, crypto forensics, compliance automation and financial crime investigation

What Is AI in Blockchain Forensics and Crypto AML?

Blockchain forensics is the process of analyzing blockchain transactions, addresses, smart contracts and asset movements to investigate suspicious or potentially illicit activity.

Crypto AML applies anti-money laundering and counter-terrorist financing controls to virtual assets and virtual asset service providers. FATF Recommendation 15 extends AML/CFT requirements to virtual assets and virtual asset service providers, while FATF continues to assess how effectively jurisdictions implement those requirements. Its 2025 targeted update said jurisdictions had made progress but continued to face important implementation challenges.

AI adds another layer to this process.

Traditional crypto compliance systems often rely on predefined rules such as:

• Wallet exposure to sanctioned entities
• Transfers involving known darknet markets
• Interactions with mixers or high-risk services
• Large or unusual transactions
• Rapid movement of assets between addresses
• Transactions involving high-risk jurisdictions

These rules remain useful, but sophisticated criminal networks can change their behavior to avoid simple detection patterns.

AI can instead analyze combinations of signals.

The important difference is that AI can look beyond an individual transaction and analyze the surrounding network.

Why Crypto AML Is Becoming More Difficult

Crypto crime is not limited to Bitcoin transactions between two wallets.

Modern illicit-finance networks can involve centralized exchanges, decentralized exchanges, bridges, stablecoins, mixers, OTC brokers, gambling services, nested services, payment processors and wallets distributed across several blockchain networks.

Chainalysis reported that illicit cryptocurrency addresses received at least $154 billion in 2025, although it emphasizes that this is a lower-bound estimate because attribution improves over time. It also reported that illicit activity remained below 1% of overall attributed crypto transaction volume and that stablecoins accounted for 84% of illicit transaction volume in its 2025 dataset.

This creates a major compliance challenge.

The result is that modern crypto AML cannot depend entirely on static rules.

Research Study 1: AI Detecting Previously Unknown Money Laundering Patterns

One of the strongest research examples comes from Elliptic and researchers associated with the MIT-IBM Watson AI Lab.

Earlier work used machine learning to identify Bitcoin transactions associated with illicit actors. The newer research expanded the approach dramatically by using a dataset containing nearly 200 million transactions.

Instead of simply asking whether an individual wallet was illicit, the model analyzed transaction subgraphs representing chains of transactions that could indicate laundering behavior.

This is important because money laundering is usually a process rather than a single transaction.

A criminal may move funds through several intermediary wallets before sending them to an exchange or another service.

The AI system attempted to identify the structure of this laundering process.

In testing with a cryptocurrency exchange, the researchers identified 52 predicted money-laundering subgraphs that ended with deposits to the exchange. The exchange confirmed that 14 involved users who had already been flagged for money laundering based on off-chain information. Elliptic reported that fewer than one in 10,000 accounts were flagged by the exchange, making the result notable as a prioritization signal.

Source: Elliptic Research, Enhancing blockchain analytics through AI

The research also identified known laundering patterns such as peeling chains and uncovered novel patterns involving intermediary services.

The most important lesson is that AI can discover patterns that were not necessarily encoded as predefined compliance rules.

This changes blockchain forensics from primarily known-pattern detection toward pattern discovery.

Research Study 2: Systematic Review of AI for Cryptocurrency Fraud Detection

A 2025 systematic review examined the development of AI-based compliance and fraud detection research in cryptocurrency transactions.

The review analyzed 353 peer-reviewed studies covering research between 2014 and 2025 using a PRISMA-based methodology.

It identified machine learning, deep learning, natural language processing and generative AI as major technology directions.

The review also identified several continuing obstacles:

• Limited transparency of AI models
• Regulatory fragmentation
• Limited access to high-quality data
• Difficulty validating models in real-world environments
• Uncertainty about long-term operational effectiveness

These findings are important because crypto AML systems operate in a high-consequence environment.

A model cannot simply produce a high accuracy score on a research dataset.

Compliance teams need to understand why an address was flagged and whether the alert can support an investigation.

Source: Finance, Systematic Review of Artificial Intelligence Applied to Compliance: Fraud Detection in Cryptocurrency Transactions, 2025

The research therefore supports a hybrid approach where AI provides prioritization and detection while investigators remain responsible for case interpretation.

Research Study 3: Systematic Review of Money Laundering in Crypto-Asset Environments

A 2026 systematic literature review examined recent research specifically focused on money laundering in crypto-asset environments.

The researchers screened 680 records and ultimately included 58 academic studies published between 2020 and 2025.

The review identified four major challenges:

The review found that graph-based and feature-based approaches were among the most important technical directions.

It also highlighted a major weakness in current research: many systems perform well in controlled environments but face difficulties with identity attribution, ground-truth labels, scalability, operational validation and cross-chain reconstruction.

Source: Financial Studies, Money Laundering in Crypto-Asset Environments: A Systematic Literature Review, 2026

This finding should strongly influence how companies build AI AML systems.

A high model score on a historical dataset is not enough.

The model must work with current blockchain behavior and integrate with real compliance investigations.

Research Study 4: AI-Based Money Laundering Detection Beyond Crypto

A 2025 systematic review in Intelligent Systems with Applications examined AI applications for money laundering detection more broadly.

The research used a PRISMA framework and analyzed AI approaches including:

• Supervised learning
• Unsupervised learning
• Deep learning
• Social network analysis
• Pattern recognition
• Other machine-learning techniques

The research highlights why graph and network methods are particularly relevant to AML.

Money laundering is fundamentally relational.

A suspicious transaction becomes more meaningful when it is considered alongside:

• Previous transactions
• Connected accounts
• Known high-risk entities
• Transaction timing
• Geographic exposure
• Asset conversion
• Intermediary services
• Repeated behavioral patterns

This supports an architecture where AI does not evaluate every transaction independently.

Instead, it evaluates the transaction’s position inside a larger network.

Source: Intelligent Systems with Applications, Review of artificial intelligence-based applications for money laundering detection, 2025

Research Study 5: AI and the Changing Crypto Crime Landscape

The 2026 crypto crime environment demonstrates why blockchain analytics needs to become increasingly adaptive.

Chainalysis reported that Chinese-language money laundering networks processed an estimated $16.1 billion in 2025, involving more than 1,799 active wallets in the network it identified.

It also estimated that these networks accounted for approximately 20% of known crypto money-laundering activity over the preceding five years.

The research identified behavioral patterns associated with different service types, including transaction fragmentation and consolidation designed to make tracing more difficult.

Source: Chainalysis, Chinese Language Money Laundering Networks, 2026

This is precisely the type of environment where AI can add value.

Criminal organizations do not necessarily repeat the same exact transaction pattern.

They can change wallets, assets, intermediaries and transaction sizes.

AI can instead look for combinations of behavioral signals.

Research Study 6: AI-Enhanced Blockchain Analytics and Unknown Wallet Discovery

Elliptic’s research provides another important finding: machine learning can help identify wallets that were not previously classified as illicit.

The model identifies suspicious transaction subgraphs first.

Investigators can then examine the wallets associated with those subgraphs and perform additional research.

This creates a discovery pipeline:

This creates a feedback loop.

Every confirmed investigation can potentially improve future detection.

That is particularly valuable in crypto because illicit actors continuously modify their infrastructure.

Source: Elliptic Research, Enhancing Blockchain Analytics Through AI

AI Blockchain Forensics Architecture

A modern crypto AML platform should combine several layers rather than relying on one model.

AI Wallet Risk Scoring

One of the most practical AI applications is dynamic wallet risk scoring.

A traditional system might assign a wallet a fixed risk category.

An AI system can continuously update the score as new information becomes available.

For example:

The system should not automatically interpret any individual signal as proof of criminal activity.

Risk signals are indicators that require context and, where appropriate, investigation.

Graph Neural Networks for Crypto AML

Graph neural networks are particularly suitable for blockchain forensics because blockchain activity naturally forms a graph.

In simplified terms:

Each wallet can be represented as a node.

Each transaction can be represented as an edge.

Additional attributes can be attached to nodes and edges:

• Transaction value
• Timestamp
• Token type
• Chain
• Service category
• Wallet age
• Transaction frequency
• Counterparty count
• Known entity labels

AI can then learn patterns in the graph.

This is more powerful than simply asking whether a single transaction looks unusual.

AI and Cross-Chain AML

Cross-chain activity is one of the biggest challenges for modern blockchain forensics.

A criminal network can move assets from one blockchain to another using bridges or swaps.

A transaction may therefore appear ordinary on one chain while forming part of a suspicious sequence when viewed across the entire ecosystem.

The AI system should build a unified identity and transaction graph.

The challenge is not simply collecting data.

The system must determine when two addresses across different networks are likely part of the same economic activity without incorrectly linking unrelated users.

This is where probabilistic entity resolution and graph-based AI become important.

AI for Stablecoin AML

Stablecoins deserve special attention because they combine crypto-native infrastructure with relatively stable monetary value.

Chainalysis reported that stablecoins represented 84% of illicit transaction volume in its 2025 analysis. The company also notes that stablecoins have broad legitimate utility because they are transferable, relatively stable and increasingly used across the crypto economy.

Source: Chainalysis, 2026 Crypto Crime Report

This means AML systems cannot simply classify stablecoin activity as suspicious.

They need contextual intelligence.

Useful AI signals include:

• Stablecoin transfer velocity
• Wallet creation patterns
• Exchange deposits and withdrawals
• Cross-chain stablecoin movement
• High-risk service exposure
• Transaction fragmentation
• Rapid conversion between assets
• Network centrality

The objective is to identify suspicious behavior while minimizing unnecessary alerts on legitimate stablecoin activity.

AI and DeFi AML Compliance

Decentralized finance creates additional challenges because there may be no traditional intermediary controlling the transaction flow.

A user can interact directly with:

• DEXs
• Lending protocols
• Bridges
• Liquidity pools
• Derivatives protocols
• Smart contracts
• Token issuers

A traditional AML platform built around centralized exchange accounts may therefore miss important parts of the transaction path.

AI can analyze protocol interactions as part of a larger graph.

For example:

The entire sequence may be more informative than any individual transaction.

AI for Suspicious Transaction Report Prioritization

One practical use of AI is not automatically filing suspicious activity reports.

Instead, AI can help compliance teams prioritize cases.

A compliance department may receive thousands of alerts.

The AI system can rank cases according to:

• Network risk
• Transaction value
• Known illicit exposure
• Behavioral similarity
• Cross-chain complexity
• Sanctions exposure
• Entity risk
• Historical case relationships

Investigators can then spend more time on the cases with stronger evidence.

This can reduce alert fatigue.

Explainable AI Is Essential for Crypto AML

A black-box AML system can create serious operational problems.

If a compliance analyst asks:

“Why was this wallet flagged?”

the system should provide understandable evidence.

A useful explanation might show:

The investigator should be able to inspect the underlying transactions.

This is more useful than simply presenting:

Risk Score: 97/100

without explanation.

Major Risks and Limitations

FATF and the Regulatory Direction

FATF continues to emphasize risk-based AML/CFT controls for virtual assets and VASPs.

Its 2025 targeted update found that implementation has improved in many jurisdictions but also identified continuing weaknesses, particularly around risk assessment and effective implementation of Recommendation 15.

Source: FATF, 2025 Targeted Update

FATF also provides guidance for assessing money-laundering risks associated with virtual assets and VASPs.

The practical implication is important:

AI should support a risk-based compliance framework rather than become the compliance framework itself.

A model can identify risk.

The organization still needs:

• Governance
• Policies
• Customer due diligence
• Transaction monitoring
• Record keeping
• Investigation procedures
• Reporting procedures
• Independent testing
• Regulatory oversight

Source: FATF, Quick Guide on Assessing Money Laundering Risks of Virtual Assets and VASPs

Expert Recommendation

The strongest approach for crypto AML is a hybrid intelligence architecture.

Companies should not attempt to replace their AML teams with one AI model.

Instead, AI should strengthen the parts of compliance where machines are particularly good at processing large amounts of structured and relational data.

The recommended architecture is:

• Use rules for known regulatory and sanctions requirements
• Use graph AI for transaction-network analysis
• Use anomaly detection for previously unknown behavior
• Use supervised learning where reliable labels exist
• Use unsupervised learning for emerging patterns
• Use NLP for external intelligence and case information
• Use entity resolution to connect on-chain and off-chain intelligence
• Use human investigators for final interpretation
• Maintain complete evidence trails for important alerts
• Continuously validate model performance

Elliptic’s 2026 outlook makes a similar broader point: technology and data are not a complete solution by themselves, but they can improve compliance efficiency, resource allocation and intelligence-led financial crime prevention.

Source: Elliptic, 2026 Regulatory and Policy Outlook

Expert Quote

A useful principle from the current blockchain analytics direction is:

The statement captures an important reality.

AI can detect patterns that humans would struggle to identify across millions of transactions, but it cannot independently determine the complete legal or factual context of every case.

The best system combines machine-scale analysis with human judgment.

AI Crypto AML Maturity Model

Implementation Roadmap

Phase 1: Build the Blockchain Data Foundation

Collect and normalize:

• Transaction data
• Wallet activity
• Token transfers
• Smart-contract interactions
• DEX transactions
• Bridge transactions
• Known service addresses
• Sanctions intelligence
• Entity information

Phase 2: Build the Transaction Graph

Represent wallets and entities as nodes and transactions as edges.

Add metadata such as:

• Transaction value
• Time
• Asset
• Chain
• Service type
• Counterparty
• Historical risk

Phase 3: Deploy Multiple AI Models

Do not depend on a single algorithm.

Use a model portfolio containing:

• Graph neural networks
• Anomaly detection
• Clustering
• Supervised classification
• Sequence models
• Entity-resolution models

Phase 4: Add Explainable Risk Scoring

Every high-risk alert should include the evidence that influenced the score.

This should allow investigators to move from:

Alert → Explanation → Transaction Path → Network → Evidence → Decision

Phase 5: Human Investigation

Investigators should be able to:

• Trace funds
• Expand transaction networks
• Compare historical behavior
• Review related entities
• Inspect cross-chain activity
• Record case decisions

Phase 6: Continuous Learning

Confirmed cases can feed back into the system.

The feedback loop becomes:

Key KPIs for AI Crypto AML

Future Predictions: 2027–2030

2027: AI Becomes a Core Investigation Layer

Crypto compliance platforms will increasingly use AI to prioritize investigations rather than relying only on static risk rules.

Investigators will receive network-level explanations instead of isolated transaction alerts.

The key change will be from:

“This transaction is suspicious”

to:

“This transaction forms part of a network pattern associated with a specific risk behavior.”

2028: Cross-Chain Intelligence Becomes Standard

As assets move between more blockchain networks, AML systems will increasingly build unified cross-chain transaction graphs.

AI will connect:

• Wallet behavior
• Bridges
• DEX activity
• Stablecoins
• Centralized exchanges
• Known service providers

The objective will be to reconstruct economic activity rather than analyze chains separately.

2029: AI Detects Emerging Laundering Typologies

Unsupervised and graph-based AI should become more important for discovering new laundering methods.

Instead of waiting for investigators to manually define a new rule, the system can identify unusual transaction structures and ask investigators to examine them.

This could shorten the time between the emergence of a new laundering technique and its addition to AML controls.

2030: Autonomous Financial Crime Intelligence Platforms

By 2030, advanced systems may combine:

The human investigator will increasingly move from manually searching millions of transactions toward reviewing machine-generated intelligence and making high-value decisions.

Startup Opportunities

AI and blockchain forensics create several opportunities for fintech, RegTech and cybersecurity companies.

• AI Crypto AML Platform for exchanges and VASPs
• Cross-Chain Forensics Engine for investigators
• AI Wallet Risk Scoring as an API
• Graph-Based Transaction Monitoring
• Stablecoin AML Monitoring
• DeFi AML Intelligence Platform
• AI Sanctions Screening
• Crypto Case Investigation Copilot
• AI SAR/STR Investigation Assistant
• Crypto Entity Resolution Platform
• AI Typology Discovery Engine
• Blockchain Financial Crime Intelligence API

One particularly valuable opportunity is an AI investigation copilot.

Such a product would not simply generate a risk score.

It could automatically reconstruct a transaction path, summarize the wallet’s behavior, identify connected services, explain the main risk signals and prepare an investigator-friendly case summary.

Frequently Asked Questions

What is AI blockchain forensics?

AI blockchain forensics uses machine learning, graph analytics and other AI techniques to analyze blockchain transactions, wallet relationships and asset flows to identify suspicious behavior and support financial crime investigations

Can AI detect crypto money laundering?

Yes, research has demonstrated that machine-learning systems can identify illicit transaction patterns and suspicious transaction subgraphs. However, AI detection is not equivalent to proving that an individual or wallet is engaged in criminal activity

Why are graph neural networks useful for crypto AML?

Blockchain transactions naturally form networks. Graph-based models can analyze relationships between wallets, services and transactions rather than evaluating each transaction independently

Can AI trace crypto across multiple blockchains?

AI can help reconstruct cross-chain activity when sufficient blockchain and entity data is available. The main challenge is accurately connecting activity across chains without creating incorrect relationships

Does blockchain transparency make crypto AML easy?

No. Public transaction data provides valuable evidence, but pseudonymity, cross-chain movement, mixers, DeFi protocols, rapidly changing wallet infrastructure and limited identity information create significant investigative challenges

Can AI replace AML investigators?

AI can automate data-intensive analysis and prioritize cases, but human investigators remain important for contextual interpretation, evidence assessment, regulatory decisions and complex investigations

What is the biggest AI opportunity in crypto AML?

One of the strongest opportunities is graph-based behavioral intelligence that can identify suspicious transaction networks and previously unknown laundering patterns rather than relying exclusively on predefined rules

Final Perspective

Crypto AML is entering a new stage.

The first generation of blockchain compliance focused heavily on address screening, transaction tracing and predefined risk rules.

Those capabilities remain important, but the scale and sophistication of modern crypto markets require a more adaptive approach.

The blockchain itself creates a powerful data environment.

Every transaction can contribute to a broader network of relationships.

Wallets interact with exchanges, bridges, DEXs, smart contracts, stablecoins and other wallets. When these interactions are combined into a graph, AI can identify structures that are difficult to recognize manually.

Recent research provides strong evidence for this direction.

Elliptic and MIT-IBM researchers demonstrated that machine learning can identify suspicious money-laundering subgraphs in a dataset containing nearly 200 million transactions and can help uncover previously unknown illicit wallets.

A 2025 systematic review covering 353 peer-reviewed studies showed the rapid growth of AI research for cryptocurrency fraud and compliance.

A 2026 systematic review of 58 crypto money-laundering studies highlighted graph-based detection as an important direction while identifying persistent challenges around labels, identity attribution, scalability and cross-chain reconstruction.

Broader AML research also shows that AI methods including supervised learning, unsupervised learning, deep learning and social-network analysis can support financial crime detection.

Meanwhile, the real-world crypto environment is changing rapidly.

Chainalysis reported at least $154 billion received by illicit cryptocurrency addresses in 2025 while emphasizing that this remains a lower-bound estimate. It also reported that illicit activity represented less than 1% of overall attributed crypto transaction volume, demonstrating why AML systems need to distinguish high-risk behavior from the overwhelmingly legitimate use of cryptoassets.

The future therefore should not be built around the assumption that every unusual transaction is criminal.

The objective should be better intelligence.

AI should help compliance teams answer:

Who is connected?

How did the assets move?

What behavior is unusual?

Does the pattern resemble known illicit activity?

Is this a new laundering typology?

What evidence supports the risk assessment?

What should the investigator examine next?

The strongest architecture will combine:

AI should not become an unexplained black box sitting between a transaction and a compliance decision.

It should become an intelligence layer that makes blockchain activity easier to understand.

The biggest transformation will therefore be:

Rule-based AML → AI-assisted AML → Graph intelligence → Cross-chain intelligence → Adaptive financial crime intelligence

For fintech companies, crypto exchanges, VASPs, banks, regulators and RegTech startups, this creates a major technology opportunity.

The winning systems will not simply flag more transactions.

They will help investigators find the right transactions, the right networks, the right entities and the right evidence faster and with greater transparency.

Research Sources

1. Elliptic Research, Enhancing Blockchain Analytics Through AI
2. Finance, A Systematic Review of Artificial Intelligence Applied to Compliance: Fraud Detection in Cryptocurrency Transactions
3. Financial Studies, Money Laundering in Crypto-Asset Environments: A Systematic Literature Review
4. Intelligent Systems with Applications, Review of Artificial Intelligence-Based Applications for Money Laundering Detection
5. Chainalysis, 2026 Crypto Crime Report
6. Chainalysis, Chinese Language Money Laundering Networks, 2026
7. FATF, Targeted Update on Implementation of the FATF Standards on Virtual Assets and VASPs, 2025
8. FATF, Quick Guide on Assessing Money Laundering Risks of Virtual Assets and VASPs
9. FATF, Virtual Assets and AML/CFT Guidance
10. Elliptic, 2026 Regulatory and Policy Outlook: The Next Generation of Blockchain Analytics
11. Elliptic, Typologies Report: AI, Crypto Crime and Emerging Financial Crime Risks