Federated Learning in AML: A Smarter Way to Fight Financial Crime in Australia

Federated learning is transforming AML by enabling banks to share intelligence without sharing sensitive data.

Introduction

Financial crime is becoming more sophisticated every year. In Australia, criminals exploit the New Payments Platform (NPP), cross-border corridors, and emerging technologies to launder billions of dollars. Banks and fintechs are under immense pressure from AUSTRAC to detect and report suspicious activity in real time.

Yet no single institution has the complete picture. Criminals spread activity across multiple banks and channels, making it difficult to detect patterns when working in isolation. This is where federated learning in AML comes in. It allows institutions to collaborate on intelligence without exposing customer data, creating a collective shield against money laundering.

What is Federated Learning in AML?

Federated learning is an artificial intelligence technique where multiple parties train a shared model without sharing their raw data. Each institution trains the model locally, and only the model updates — not the underlying data — are shared.

In AML, this means:

• Banks contribute insights into suspicious patterns.
• Sensitive customer data remains within each institution.
• A shared model learns from multiple perspectives, strengthening detection.

It is compliance collaboration without compromising privacy.

Why Australia Needs Federated Learning

1. Fragmented Data

Each bank only sees part of the financial ecosystem. Criminals exploit these gaps by spreading transactions across multiple institutions.

2. Rising Compliance Costs

Institutions are spending billions annually on AML compliance. Shared learning reduces duplication of effort.

3. AUSTRAC’s Push for Innovation

AUSTRAC encourages industry collaboration to strengthen financial crime prevention. Federated learning aligns perfectly with this goal.

4. Real-Time Payment Risks

With NPP and PayTo, money moves instantly. Federated learning enables faster identification of emerging fraud typologies.

5. Protecting Privacy

Australia’s data protection regulations make raw data sharing complex. Federated learning solves this by keeping sensitive data local.

How Federated Learning Works in AML

1. Local Training
   Each institution trains an AI model on its transaction and customer data.
2. Model Updates Shared
   Only the learned patterns (model weights) are sent to a central aggregator.
3. Global Model Improved
   The aggregator combines updates from all banks into a stronger model.
4. Distribution Back to Banks
   The improved model is sent back to each bank for use in detection.

This cycle repeats, continually improving AML detection across the industry.

Use Cases of Federated Learning in AML

1. Mule Account Detection
   Identifies networks of mule accounts across different banks.
2. Cross-Border Laundering
   Tracks layering activity spread across institutions and jurisdictions.
3. Fraud Typology Sharing
   Allows banks to learn from each other’s fraud cases without sharing customer data.
4. Sanctions Screening Enhancement
   Improves detection of high-risk entities that use aliases or complex networks.
5. Customer Risk Profiling
   Builds more accurate risk scores by learning from industry-wide patterns.

Benefits of Federated Learning in AML

• Collective Intelligence: Stronger models built from multiple perspectives.
• Privacy Protection: Raw customer data never leaves the institution.
• Faster Adaptation: New fraud typologies shared quickly across banks.
• Cost Efficiency: Reduces duplication of AML technology spend.
• Regulatory Alignment: Demonstrates proactive industry collaboration.

Challenges of Federated Learning

• Data Quality: Poor-quality local data reduces model accuracy.
• Technical Complexity: Requires strong IT infrastructure for secure collaboration.
• Coordination Barriers: Banks must align on frameworks and standards.
• Explainability: AI models must remain transparent for AUSTRAC compliance.
• Adoption Costs: Initial investment can be high for smaller institutions.

Spotlight: Tookitaki’s AFC Ecosystem and FinCense

Tookitaki has pioneered federated learning in AML through its AFC Ecosystem and FinCense platform.

• AFC Ecosystem: A global community of compliance experts contributing real-world scenarios and typologies.
• Federated Learning Engine: Allows banks to benefit from collective intelligence without sharing raw data.
• Real-Time Monitoring: Detects suspicious activity across NPP, PayTo, remittance corridors, and crypto.
• FinMate AI Copilot: Assists investigators with summarised alerts and regulator-ready reports.
• AUSTRAC-Ready: Generates SMRs, TTRs, and IFTIs with full audit trails.
• Cross-Channel Coverage: Unifies detection across banking, wallets, cards, remittances, and crypto.

By combining federated learning with Agentic AI, FinCense delivers industry-leading AML capabilities tailored for the Australian market.

Best Practices for Adopting Federated Learning in AML

1. Start with Partnerships: Collaborate with trusted peers to test federated models.
2. Focus on Data Quality: Ensure local models are trained on clean, structured data.
3. Adopt Explainable AI: Maintain regulator confidence by making outputs transparent.
4. Engage Regulators Early: Keep AUSTRAC informed of federated learning initiatives.
5. Invest in Infrastructure: Secure, scalable platforms are essential for success.

The Future of Federated Learning in AML

1. Industry-Wide Collaboration: More banks will join federated networks to share intelligence.
2. Real-Time Typology Sharing: Federated systems will distribute new fraud scenarios instantly.
3. Cross-Sector Expansion: Insurers, payment firms, and fintechs will join federated AML networks.
4. Global Interoperability: Federated learning models will connect across borders.
5. AI-First Investigations: AI copilots will use federated intelligence to guide case investigations.

Conclusion

Federated learning in AML represents a breakthrough in the fight against financial crime. By combining intelligence from multiple banks without exposing customer data, it creates a collective defence that criminals cannot easily evade.

In Australia, where AUSTRAC demands stronger monitoring and fraudsters exploit instant payments, federated learning provides a powerful solution.

Platforms like Tookitaki’s FinCense are making federated learning a reality, turning compliance from a siloed burden into a shared advantage.

Pro tip: The future of AML will be built on collaboration. Federated learning is the foundation that makes industry-wide intelligence sharing possible.