Feedzai Launches RiskFM, the First Tabular Foundation Model for Financial Crime Detection

A new class of AI for the financial‑crime battlefield

On March 24, 2026, Feedzai – the AI native leader in financial‑crime prevention – announced the release of RiskFM (Risk Foundation Model), positioning it as the industry’s inaugural foundation model built specifically for tabular financial data. The announcement, issued from New York and Lisbon, marks a departure from the rule‑based and bespoke machine learning approaches that have dominated fraud and anti‑money‑laundering (AML) defenses for decades.

RiskFM is billed as a “purpose‑built frontier model” that can be applied across the entire risk‑decision lifecycle: from onboarding and digital activity monitoring to real‑time payments, transfers and AML investigations. Feedzai claims the model is trained on a uniquely extensive, global dataset that spans more than $9 trillion in annual payments and 120 billion events, covering a full spectrum of financial interactions.

Why “tabular” matters – and why it’s hard

Large language models (LLMs) have reshaped natural‑language processing, image recognition and video analysis because those domains are bounded by relatively fixed grammars and spatial continuity. In contrast, transactional data is highly heterogeneous, constantly evolving, and deliberately obfuscated by adversaries. As Feedzai’s chief science officer Pedro Bizarro put it, “Next transactions are far less predictable than the next word in a sentence. 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.”

Tabular data – the rows and columns that record every payment, login, or transfer – lacks the clear sequential or visual structure that LLMs exploit. Building a foundation model that can learn from billions of such records without overfitting to a single institution’s quirks is therefore a technical leap.

Feedzai’s data moat

Feedzai’s claim of training RiskFM on a “uniquely broad, deep, global dataset” is more than marketing speak. The company processes risk assessments on $9 trillion of payments each year, aggregating data from 120 billion events that cut across onboarding, digital activity, card payments, real‑time transfers and AML workflows. This breadth allows RiskFM to ingest patterns from multiple geographies and product lines, providing what Feedzai calls “compounding intelligence.”

When the model is trained on data from several institutions simultaneously, Feedzai reports that it “outperforms traditional models based on Gradient Boosting and Deep Learning approaches, and keeps improving as it ingests more data.” The implication is that the model benefits from cross‑institutional signal sharing, a capability that most proprietary fraud engines lack.

Performance claims: matching and surpassing bespoke solutions

According to the press release, RiskFM can “match the performance of highly‑tuned supervised models on Day One for a single customer, without the manual feature engineering that usually consumes weeks of data‑science effort.” Moreover, when the model is applied across multiple customers, it “surpasses” those bespoke solutions, delivering “more value for customers, faster deployment times, and significantly lower implementation and maintenance costs.”

These assertions are reinforced by an IDC commentary. Sam Abadir, research director for risk, financial crime and compliance at IDC, noted, “Foundation models have reshaped language, vision, and audio, but financial crime has remained stubbornly resistant to that wave. 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.”

A single model for the whole risk stack

RiskFM is presented as a unified AI layer capable of handling diverse use cases, from mule‑account detection to AML screening. Feedzai lists three core capabilities:

• Compounding intelligence – the model improves as it learns from additional institutions and regions.
• Day‑One parity – it delivers performance comparable to highly‑engineered, customer‑specific models without bespoke feature work.
• End‑to‑end applicability – a single foundation can serve multiple risk functions, reducing the need for separate models for fraud, AML, and related controls.

Pedro Barata, Feedzai’s chief product officer, emphasized the strategic intent: “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.”

Early adopters and collaborative validation

Feedzai says it is already working with “early adopters” to validate the initial RiskFM frameworks, with plans to expand the methodology to larger datasets and integrate the model across its full suite of solutions. One notable collaborator is Lloyds Banking Group. Tom Martin, Lloyds’ Business Platform Lead for Economic Crime Prevention, remarked, “Lloyds Banking Group works collaboratively across the industry to protect consumers from financial crime. 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.”

The partnership underscores a broader industry trend: banks are increasingly seeking AI tools that can be deployed rapidly while satisfying stringent regulatory expectations.

Regulatory backdrop and compliance considerations

Financial institutions operate under a patchwork of AML, sanctions and consumer‑protection regulations that demand both effectiveness and explainability. While foundation models excel at pattern recognition, they can be opaque. Feedzai’s claim of “no manual feature engineering” raises questions about model interpretability and auditability. Regulators such as the FCA and the European Banking Authority have signaled a willingness to accept AI‑driven controls, provided firms can demonstrate robust governance, risk‑assessment and documentation.

If RiskFM can indeed deliver “Day‑One” performance without extensive feature‑engineering, firms will need to ensure that the model’s decision logic can be traced, that false‑positive rates stay within acceptable limits, and that the model can be updated in response to emerging typologies without violating model‑risk management frameworks.

Competitive landscape: where does RiskFM fit?

The market for AI‑enabled fraud and AML solutions is crowded. Established players like FICO, SAS and Nice Actimize offer rule‑based engines and custom machine‑learning pipelines, while newer entrants such as Darktrace and Forter focus on unsupervised anomaly detection. Most of these solutions rely on institution‑specific training data, limiting their ability to benefit from cross‑industry learning.

RiskFM’s tabular foundation approach could give Feedzai a differentiator, especially if the model’s “compounding intelligence” delivers measurable lift over siloed models. However, competitors are also exploring large‑scale pre‑training on transactional data. The success of RiskFM will hinge on Feedzai’s ability to continuously source fresh, high‑quality data, maintain compliance with data‑privacy laws (e.g., GDPR), and prove the model’s ROI in real‑world deployments.

Business impact and strategic positioning

From a business perspective, the launch signals Feedzai’s ambition to move beyond a vendor of point solutions toward a platform provider that can serve the entire risk ecosystem. By offering a single model that can be fine‑tuned for fraud, AML, and broader risk decisions, Feedzai may reduce integration complexity for clients and lock in longer‑term relationships.

If the performance claims hold, institutions could see faster time‑to‑value, lower engineering overhead and a more agile response to emerging fraud typologies. For investors, the move could be interpreted as a signal of Feedzai’s confidence in its data moat and its capacity to monetize foundation‑model technology across a broader addressable market.

Outlook: a foundation model for a volatile risk environment

The financial‑crime landscape is in constant flux, driven by faster payment rails, the rise of digital‑only banking and increasingly sophisticated criminal networks. Traditional rule‑based systems struggle to keep pace, while bespoke machine‑learning models demand significant data‑science resources.

RiskFM, as a tabular foundation model, aims to bridge that gap by offering a reusable, continuously improving AI core. Its success will depend on three factors: (1) the ability to maintain high detection rates while keeping false positives low, (2) regulatory acceptance of a less‑transparent but more powerful model, and (3) the scalability of Feedzai’s data pipeline to ingest ever‑growing transaction volumes.

If these hurdles are cleared, RiskFM could redefine how banks, payment processors and fintechs approach risk, ushering in an era where a single AI engine underpins fraud detection, AML screening and broader risk‑management functions.

Get in touch with our fintech experts