Feedzai Unveils RiskFM AI Foundation Model for Financial Crime Prevention

WHY THIS MATTERS: The launch of a dedicated foundation model for financial data signals a paradigm shift in the battle against sophisticated financial crime. For years, institutions have struggled with a fragmented defense landscape, where fraud detection and anti-money laundering (AML) operate on disparate, manually-engineered systems. This new, holistic approach leverages vast global datasets across the entire risk lifecycle—from customer onboarding to transaction monitoring—to create a singular, adaptive intelligence layer. In the age of instant transactions, where a successful fraud attack must be stopped in milliseconds, relying on legacy, single-purpose models is a liability. This development promises to significantly reduce the cost and deployment time required for banks to achieve comprehensive financial crime prevention, making advanced AI defenses accessible and scalable across complex global operations.

Feedzai, the global leader in AI-native financial crime prevention, today unveiled RiskFM (Risk Foundation Model), the industry’s first Tabular Foundation Model purpose-built for financial data and risk decisioning.

RiskFM marks a fundamental shift in how financial crime is detected and prevented. For decades, institutions have relied on rules and manually-engineered machine learning models built one customer at a time. RiskFM changes that as a purpose built frontier model that spans across fraud detection, anti-money laundering (AML), and broader risk decisions across the entire financial crime lifecycle. Unlike current industry attempts limited to card network data, RiskFM is trained on a uniquely broad, deep, global dataset spanning onboarding, digital activity, payments, transfers, and AML workflows, enabling institutions to detect, prevent, and adapt to financial crime with unprecedented speed and precision.

Solving the Unique Challenge of Transactional Data

Recently, Large Language Models (LLMs) have effectively “solved” domains like language, audio, and video because they are highly constrained by finite grammar and causality. In language, next words are often predictable: in the sentence “Yesterday, a scammer contacted me and pretended to be my …”, the next word is likely “relative,” “friend,” or “coworker”. Similarly, in images and video, individual pixels are highly predictive of their nearby neighbors. Financial transactions, however, operate in a fundamentally different reality.

“Next transactions are far less predictable than the next word in a sentence,” said Pedro Bizarro, chief science officer at Feedzai. “Consumer spending habits, payment types, and fraud modes change continuously. More importantly, financial risk is an adversarial domain; fraudsters actively adapt to evade detection in real-time.”

Feedzai is uniquely positioned to explore large datasets, as the company annually risk-assesses $9T in payments across 120B events worldwide that span the entire financial risk lifecycle: from onboarding and digital activity to card payments and real-time transfers. This unparalleled breadth ensures RiskFM is tested at scale as a holistic model, rather than siloed in a single specialized application.

RiskFM is already showing it can match the performance of bespoke supervised models even with data from a single customer, and it surpasses them when trained with data from several institutions and geographies. The result is more value for customers, faster deployment times, and significantly lower implementation and maintenance costs.

“Foundation models have reshaped language, vision, and audio, but financial crime has remained stubbornly resistant to that wave,” said Sam Abadir, research director, risk, financial crime, and compliance for IDC. “Feedzai’s RiskFM is a credible attempt to close that gap. The ability to match bespoke supervised models out of the box, without manual feature engineering, has real implications for how institutions think about deployment speed, cost, and coverage across the full financial crime lifecycle, from card fraud to AML. The early performance data is worth watching, as is how the model holds up as it expands into more complex use cases.”

A Unified Model With Unprecedented Performance

Following rigorous testing and baseline experiments, RiskFM delivers unprecedented capabilities:

• Compounding intelligence: When trained across multiple institutions and geographies simultaneously, RiskFM outperforms traditional models based on Gradient Boosting and Deep Learning approaches, and keeps improving as it ingests more data.
• Ability to match highly-tuned models on Day One: When RiskFM is used to power a bespoke model for a single customer, it matches the performance of high-tuned supervised models without manual, time-consuming feature engineering.
• One model from mule account detection to AML: RiskFM serves as the foundational AI layer for financial risk. It is designed to expand across the full range of financial crime prevention, from mule account detection to AML, providing institutions with a scalable, intelligent model that grows with their needs.

“Our vision is coming true: this is not just another Large Tabular Model for a single data type. We’ve developed a foundation model for financial data that covers multiple use cases — from cards to real-time payments — and geographies, delivering strong performance from Day One at global scale,” said Pedro Barata, chief product officer at Feedzai. “RiskFM proves our multi-year investment in foundation models is paying off. We’re not just part of the conversation; we’re defining how it applies to the complexities of global financial crime prevention.”

Feedzai is working with early adopters to validate initial RiskFM frameworks and plans to scale these methodologies to large datasets, ultimately integrating them across its full suite of use cases.

“Lloyds Banking Group works collaboratively across the industry to protect consumers from financial crime,” said Tom Martin, Lloyds Banking Group Business Platform Lead, Economic Crime Prevention. “We’ve been collaborating with Feedzai for years on AI innovation to give fraud fighters the upper hand against criminals, and RiskFM is an exciting milestone in that journey.”

FF NEWS TAKE: This is a decisive move that validates the next generation of AI in finance, shifting the focus from discrete machine learning models to enterprise-wide foundation models. It critically moves the needle by offering a unified approach to two historically separate functions: fraud and anti-money laundering. The key test now is not the initial accuracy, but governance: can regulators and global institutions quickly adapt compliance frameworks to handle a shared, compounding intelligence layer? Watch for rival vendors to quickly follow suit with their own unified risk decisioning platforms.