Breaking AML Barriers with Federated Learning: From Policy to Practic

For years, the fight against money laundering has been described as a “shared problem in need of a shared solution.” Regulators, financial intelligence units (FIUs), and banks all agree that no single institution can combat financial crime alone. Yet despite the consensus, cross-institutional collaboration in anti-money laundering (AML) often stalls before moving from policy papers to operational reality.

Why? Because collaboration in AML is harder than it looks. Regulatory fragmentation, data privacy concerns, legacy infrastructure, and even competitive pressures mean that well-intentioned initiatives frequently stall or collapse. The challenge isn’t a lack of ambition—it’s the barriers that prevent information from being shared and acted upon securely.

But a promising shift is emerging: federated learning (FL). This approach enables financial institutions to train machine learning models across multiple data sources without moving or exposing sensitive information. And for the first time, it offers a practical way to bridge the AML collaboration gap.

Why Traditional AML Collaboration Struggles

Several high-profile projects have demonstrated both the potential and pitfalls of AML collaboration. The UK’s National Crime Agency (NCA) Data Fusion programme has shown that when regulators and institutions focus on high-impact use cases, collective intelligence can deliver real results. Similarly, the Netherlands’ Transaction Monitoring Netherlands (TMNL) pilot demonstrated the ability of banks to pool data to detect suspicious behaviour at scale.

But these projects are exceptions. Many others never progress beyond pilots. Consider:

• Regulatory misalignment: FATF may set global AML/CFT standards, but each jurisdiction interprets and enforces those standards differently. A compliance model that works in London may not hold in Jakarta or Mumbai.

  
  

• Data privacy risks: Regulations such as GDPR in Europe or Singapore’s PDPA impose strict requirements on data storage and usage. Even anonymised data can present compliance and reputational risks.

  
  

• Siloed institutions: Within banks themselves, legal, compliance, and technology teams often work in isolation, limiting their ability to share intelligence—even internally across borders.

  
  

• Technology gaps: Legacy infrastructure, incompatible formats, and limited interoperability make data sharing cumbersome and costly.

  
  

The result? Many collaborative AML projects fall victim to operational friction, governance disputes, or shifting regulatory requirements.

Lessons from Asia-Pacific

The Asia-Pacific (APAC) region illustrates both the challenges and the opportunities in AML collaboration.

In Singapore, the Monetary Authority of Singapore’s AML/CFT Industry Partnership (ACIP) has been widely regarded as a success. By focusing on specific typologies—such as trade-based money laundering or shell company misuse—banks, regulators, and law enforcement created a targeted framework for intelligence sharing. Crucially, MAS provided legal clarity and governance oversight, reducing the risk for participants.

Contrast this with efforts in other markets. In Australia, the absence of a unified AML data-sharing model means financial institutions often duplicate efforts when detecting cross-institutional risks. In Hong Kong, strict data residency requirements make even intra-group sharing across borders difficult. Meanwhile, in Indonesia and the Philippines, varied interpretations of FATF recommendations add further complexity for multinational institutions.

These realities highlight a clear truth: AML collaboration only works when it is private by design, aligned with local regulation, and operationally feasible.

Federated Learning: A Breakthrough for AML

Federated learning offers a way to overcome these barriers. Instead of pooling raw data in a central hub, FL allows institutions to train models locally on their own datasets. Only encrypted model updates—not sensitive customer data—are shared across participants.

The advantages are significant:

• Data stays local: Institutions maintain full control over their data, reducing privacy and compliance risks.

  
  

• Collective intelligence: By pooling insights rather than raw records, FL captures behavioural patterns that no single bank could detect in isolation.

  
  

• Regulatory alignment: Because data never leaves the originating institution, FL is better suited to comply with strict data residency and privacy laws.

  
  

• Operational efficiency: Early pilots have reported up to a fourfold uplift in detection rates and a 75% improvement in analyst productivity, as models learn from a richer set of behaviours while reducing false positives.

  
  

Practical Applications in APAC

Imagine a scenario in Malaysia, where regulators have intensified their focus on mule accounts used in scam networks. With federated learning, local banks could train detection models on their own transaction patterns while benefiting from collective updates across multiple institutions. No customer data crosses borders, yet each participant benefits from a broader intelligence pool.

Or consider cross-border remittance flows in the Philippines and Vietnam. These corridors are vital for millions of households but often exploited by criminal networks layering illicit funds. Federated learning could allow remittance providers and banks to jointly spot unusual transfer behaviours while complying with local privacy frameworks.

Even in Australia’s gaming sector, where regulators are tightening scrutiny of casinos and digital betting platforms, federated learning could support operators in detecting suspicious betting or payout patterns that span across institutions.

Moving from Intent to Action

The lesson from past efforts is clear: collaboration in AML fails when it is too broad, too exposed, or too uncertain. Federated learning, however, is private by design, outcome-driven, and technically feasible. It allows institutions to share strength without sharing secrets.

For the APAC region—home to some of the world’s fastest-growing financial markets and most diverse regulatory environments—the potential is particularly powerful. With federated learning, collaboration need not mean compromise. Instead, it can finally move from good intentions and policy pilots to measurable impact.

The barriers to AML collaboration are real, but they are no longer insurmountable. Federated learning provides the bridge—turning shared intent into shared action.