The opinions expressed here are those of the authors. They do not necessarily reflect the views or positions of UK Finance or its members.

2025 is going to be the year to focus on fraud. The government's message is clear, with initiatives like the UK’s Economic Crime Plan 2.0, which prioritises anti-fraud measures, and the EU’s ongoing efforts to enhance financial security. Regulators are also increasing their activity, as seen in the European Banking Authority’s (EBA) enhanced fraud reporting requirements and the FCA’s push for stricter cybersecurity and fraud disclosure regulations.

The urgency for better fraud prevention is not just regulatory-driven—it is a business necessity. The global cost of financial fraud is expected to surpass £500 billion annually, with cybercriminals leveraging increasingly sophisticated techniques, including deepfake social engineering, synthetic identities, and AI-powered money laundering.

Financial institutions are responding with significant investments in AI-driven fraud detection, with major banks such as HSBC and Barclays leading the way. But despite these advances, fraudsters continue to exploit gaps between institutions, operating across multiple banks, fintechs, and payment processors.

The challenge remains: How can financial institutions collaborate effectively on fraud prevention without compromising customer data privacy?

Why financial institutions struggle with data sharing

As you know, financial criminals do not target a single bank—they move between financial institutions, exploiting weak points across the sector. This means fraud detection is most effective when organisations share intelligence. However, privacy, compliance, and reputational risks have made widespread data sharing difficult.

Initiatives such as industry-wide fraud databases have traditionally attempted to bridge this gap. However, concerns over GDPR compliance, competitive sensitivity, and consumer trust have often prevented institutions from fully embracing collaborative fraud detection models.

Until now…

Federated learning: A privacy-first approach

Federated learning offers financial institutions a way to strengthen fraud detection without sharing raw data. This machine learning technique enables multiple organisations to train fraud detection models collectively, ensuring insights are shared while keeping sensitive customer data private.

At the recent UK Finance Economic Crime Congress, federated learning was highlighted as a critical innovation in financial crime prevention. It allows financial institutions to detect fraud patterns earlier by learning from trends across multiple organisations—without ever pooling or exposing customer data.

How federated learning works

Federated learning enables banks and financial institutions to improve fraud detection through a collaborative but secure model:

1. Local model training: Each institution trains a fraud detection model using its own customer data.
2. Secure model aggregation: Instead of sharing raw data, only encrypted model updates (patterns and insights) are sent to a central aggregator.
3. Global model improvement: The aggregator refines the model using insights from all institutions and shares an improved version back to participants.
4. Continuous learning: This cycle repeats, ensuring fraud detection capabilities evolve with new threats.

This privacy-preserving approach allows institutions to benefit from collective intelligence while remaining compliant with GDPR and data protection regulations.

The future of fraud prevention

Collaboration will be essential for federated learning to become a core fraud prevention tool in 2025. Banks, fintech, regulators, and technology providers must collaborate to build scalable, privacy-secure implementation frameworks.

Advances in privacy-enhancing technologies, such as differential privacy and secure multiparty computation, will further improve the security and trustworthiness of federated learning-based fraud prevention systems. As fraud threats become more complex, financial institutions must embrace new ways to detect, share, and act on fraud intelligence—while protecting customer data.

Final takeaway: Federated learning represents a significant step forward in the fight against fraud. By enabling institutions to collaborate without compromising privacy, it offers a way to outpace fraudsters while maintaining compliance. In 2025, financial institutions must prioritise secure, data-driven collaboration.