November 17, 2026 · 5min read

Federated Learning: Sharing Fraud Intelligence Without Sharing Data

Bruno Terroso
Bruno Terroso

Product Marketing Specialist

Bruno Terroso is part of Feedzai's Product Marketing team, creating content that helps financial institutions fight fraud and financial crime.

Why is Cross-border Fraud Prevention Collaboration Still Do Difficult?

What Intelligence-sharing Practices is the Industry Trying?

How Does Federated Learning Work?

Taking Federated Learning from Pilot to Production

What Happens Next for Collaboration and Federated Learning?

FAQs about Federated Learning and Fraud Intelligence Sharing

At the Global Fraud Summit earlier this year, Financial Action Task Force and Interpol leaders highlighted that the global fraud economy now accounts for over $1 trillion in illicit financial flows every single year, a figure first revealed by the Global Anti-Scam Alliance.1 

This trillion-dollar figure demonstrates that the stakes for fraud prevention leaders have never been higher. Today’s scams are highly networked, meaning that an attack that starts at one bank can cross account boundaries, payment rails, and borders within hours. The fatal limitation of most traditional defenses is that they are siloed, focusing only on what a single institution can see. Fortunately, financial institutions have a solution within their reach: federated learning. Federated learning enables organizations to share vital fraud intelligence across networks without compromising data or undermining customer privacy.

The Future of Financial Intelligence Sharing (FFIS) program at RUSI recently looked at how the industry is trying to change that. Its May 2026 report, A Survey of Collaborative Analytics Platforms to Tackle Fraud: Towards a Network of Networks, examines 37 platforms and partnerships worldwide to understand how these collaborative analytics platforms share and analyze operational data.2

One finding in the FFIS report stands out. While 91% of the platforms surveyed consider cross-border fraud-risk sharing essential, most report little or no meaningful cooperation with platforms in other countries.

Key Takeaways

  • Global fraud accounts for over $1 trillion in illicit financial flows annually; a recent FFIS report highlights the urgent need for collaborative defense strategies across organizations and borders.2
  • Despite 91% of platforms considering cross-border fraud-risk sharing essential, intelligence sharing is often hindered by strict data protection rules, varying fraud taxonomies across jurisdictions, and a lack of international standards.
  • Federated learning enables institutions to train models locally and share collective insights without moving sensitive raw data or sacrificing real-time, low-latency performance.
  • Feedzai IQ Score, already deployed across multiple countries, utilizes federated learning to deliver real-time risk scores without compromising data privacy.

Why is Cross-border Fraud Prevention Collaboration Still Do Difficult?

There are obvious reasons organizations hesitate to collaborate on sharing intelligence. Data protection rules can restrict how information moves between institutions, particularly across borders. Different jurisdictions also use different fraud taxonomies and data formats. Plus, no international body sets common standards.

Then there is the speed question.

The most commonly cited challenge in the FFIS survey, mentioned by 64% of platforms, is the speed and diversity of modern payments. One respondent described real-time payments as a “force multiplier for fraud and scam syndicates.”

Feedzai IQ Solution Guide

Financial institutions often build fraud defenses based on their own environment: their own customers, their own transaction history, and their […]

Learn More

The key takeaway? Intelligence is only useful if it arrives in time to act. If the money has already moved, understanding what happened afterward doesn’t help much.

This shows the challenge isn’t simply sharing more intelligence. It’s finding ways to make that intelligence useful without exposing customer data or adding unnecessary friction to decisions that may need to happen in milliseconds.

What Intelligence-sharing Practices is the Industry Trying?

The FFIS report covers a broad range of collaborative approaches, each solving a different part of the problem.

  • Shared warning lists. Members contribute known bad actors, and everyone checks against the pool. Salv Bridge, for example, lets banks screen new applicants against a shared, cross-border list of high-risk and offboarded customers.
  • Secure messaging and joint investigations. Institutions work a specific case together. Under Section 314(b) in the US, a group of major banks reports identifying an average of four additional suspects for every subject law enforcement shares.
  • Real-time tracing. When stolen funds move through mule accounts, every institution in the chain is alerted. 
  • Privacy-enhancing technologies. These answer a question without revealing the underlying data. Roseman Labs uses multi-party computation, so three large European banks can match against each other’s fraud watchlists without disclosing them. It’s promising, but still at the pre-production stage.

Each approach has its place: warning lists and messaging for known threats and cases, tracing for following stolen funds, encrypted matching for narrow questions.

Federated learning addresses a different challenge. It lets institutions learn from network-wide patterns without moving their underlying data, while still delivering the ultra-low latency modern fraud prevention requires.

How Does Federated Learning Work?

The idea behind federated learning is fairly simple.  Instead of sending all the data to a central model, each institution trains a model using its own data in its own environment. The network can then combine the resulting model knowledge.

Customer data stays where it is while learning travels across the network. FFIS’ report describes this as participants “pooling their insights” to create models that can identify a broader range of fraud patterns than any individual institution might see on its own.

As with any other approach, there are trade-offs. For example, data quality and fraud labels can vary between participants, and keeping model updates aligned across institutions requires technical work. Additionally, federated learning isn’t designed to map mule networks between institutions because it doesn’t exchange operational data. Other types of intelligence-sharing capabilities are better suited for that task.

Federated learning isn’t a replacement for every other form of collaboration; its value is more specific: it lets institutions learn from a broader network while keeping sensitive data in their own environments.

Taking Federated Learning from Pilot to Production

One thing the FFIS report notes is that some of these collaborative approaches are still in early stages.

Several of the privacy-preserving technologies described, including federated learning for fraud labels, are being tested through pilots and sandbox programs. Meanwhile, Feedzai IQ Score is already deployed across multiple countries.

Feedzai IQ uses federated learning to analyze metadata from participating members, including members operating in different countries. The network uses these insights to develop models and risk signals that improve fraud detection. Personal data remains within each client’s PCI DSS-compliant environment. Then, it delivers risk scores and risk factors for each transaction in real time.

Preliminary results from live deployments found that one merchant acquirer saw a 27% improvement in payment acceptance rates and a 5% increase in fraud detection after adding Feedzai IQ.

These results are not universal guarantees. Outcomes depend on factors such as an institution’s size and geography. The report notes that platform results are self-reported, but they illustrate the potential: better fraud protection without unnecessarily declining good customers.

What Happens Next for Collaboration and Federated Learning?

The FFIS report makes a case for a “network of networks” and recommends that the FATF lead the development of cross-border standards, legal gateways, and common data protocols.

That work matters. Common standards would make it much easier for institutions and existing networks to work together. However, policy development takes time while fraud doesn’t.

Technologies such as federated learning already enable collaboration while keeping sensitive data local, helping institutions learn from a much broader pool of intelligence while still operating within their existing data environments and at the speed required by modern payments.

We’re proud to support the FFIS Innovation Corridors and to contribute to this broader effort to make financial intelligence more collaborative.

Additional Resources

FAQs about Federated Learning and Fraud Intelligence Sharing

What is fraud intelligence sharing?

Fraud intelligence sharing is the process where financial institutions, payment networks, and risk platforms exchange information about fraud patterns, tactics, and suspect accounts. Instead of relying solely on isolated internal data, organizations collaborate to gain a broader view of emerging threats across the ecosystem, helping prevent cross-border scams and coordinated attacks before significant financial damage occurs.

How can banks share fraud intelligence without sharing customer data?

Banks can use privacy-enhancing technologies like federated learning, multi-party computation, and encrypted matching. These technologies allow institutions to train shared analytical models or verify specific fraud signals using local data or anonymized metadata. As a result, sensitive personally identifiable information (PII) stays within each bank’s secure environment, ensuring regulatory compliance and safeguarding privacy.

What is federated learning in fraud prevention?

Federated learning is a decentralized machine learning approach where algorithms are trained locally across multiple institutions on their own private data. Instead of centralizing raw customer information, only model updates and insights travel across the network to build a collective intelligence. This allows banks to benefit from network-wide fraud insights while maintaining strict local data protection.

How does federated learning help financial institutions detect fraud?

It improves risk scoring by training models on broader, network-wide fraud patterns that no single institution could observe alone. By uncovering complex cross-bank attack schemes in real time without introducing heavy latency, federated learning boosts fraud detection accuracy, increases legitimate transaction approval rates, and reduces false positives without compromising customer privacy or exposing sensitive operational data.

What are the main challenges of cross-border fraud intelligence sharing?

Key challenges include strict cross-border data privacy regulations, inconsistent legal frameworks, diverse fraud taxonomies, and the lack of international standardizing bodies. Additionally, modern real-time payment systems demand ultra-low-latency decisioning, making traditional intelligence-sharing methods too slow. Technologies like federated learning help bypass these obstacles by enabling instant collaborative insights without physically moving sensitive data across borders.

Footnotes

1 https://fincrimecentral.com/fatf-interpol-global-strategy-fraud-recovery/#:~:text=Cooperation%20between%20the%20public%20and%20private%20sectors,INTERPOL%20Financial%20Crime%20and%20Anti%2DCorruption%20Centre%20Resources

2 https://www.future-fis.com/fraudplatforms.html

All expertise and insights are from human Feedzaians, but we may leverage AI to enhance phrasing or efficiency. Welcome to the future.

Page printed in November 19, 2026. Please see https://www.feedzai.com/blog/federated-learning-shared-fraud-intelligence for the latest version.