What AI builders can learn from fraud models that run in 300 milliseconds

2026-02-09 · Source: VentureBeat · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Data Science & Analytics · Depth: Intermediate, medium

Summary

Mastercard's Decision Intelligence Pro (DI Pro) is an AI-powered fraud detection platform designed to analyze individual transactions in milliseconds. The system processes approximately 160 billion transactions annually, with surges up to 70,000 transactions per second during peak times. DI Pro utilizes a recurrent neural network (RNN) with an "inverse recommender" architecture, treating fraud detection as a pattern completion problem to assess transaction risk. This architecture identifies how merchants relate to one another and evaluates if a transaction aligns with a user's typical behavior. Mastercard also employs techniques like aggregated, anonymized data to maintain data sovereignty while leveraging global fraud patterns, and uses "honeypots" to engage cybercriminals and map global fraud networks by identifying mule accounts.

Key takeaway

For Machine Learning Engineers developing high-throughput, low-latency systems, consider adopting an "inverse recommender" RNN architecture for real-time risk assessment, similar to Mastercard's DI Pro. Your focus should be on optimizing models for sub-300ms inference times and integrating data sovereignty solutions like anonymized global patterns to enhance decision quality without compromising privacy. Prioritize projects with strong business impact and ensure thorough activation before full implementation.

Key insights

AI models can detect fraud in milliseconds by analyzing individual transaction patterns and user behavior.

Principles

• Fraud detection benefits from real-time, contextualized risk scoring.
• Data sovereignty can be maintained using aggregated, anonymized data.
• Proactive engagement with fraudsters can map global networks.

Method

DI Pro uses an "inverse recommender" RNN to perform pattern completion on transaction data, assessing if a merchant interaction makes sense for a user based on past behavior. It also employs honeypots to identify mule accounts.

In practice

• Implement RNNs for real-time anomaly detection in high-volume systems.
• Utilize anonymized data for global pattern analysis while respecting data sovereignty.
• Deploy honeypots to gather intelligence on fraudster networks.

Topics

• Fraud Detection AI
• Recurrent Neural Networks
• Real-time Transaction Processing
• Data Sovereignty
• Cybercrime Defense

Best for: Machine Learning Engineer, AI Engineer, Data Scientist, AI Product Manager

Related on AIssential

• How Stripe Detects Fraudulent Transactions Within 100 ms — Evolving ML architectures and leveraging vast network data are crucial for high-performance fraud detection.
• Beyond Defensive Reporting: Machine Learning for Active Anti-Money Laundering Control in Insurance — Machine learning, particularly with fraud data, can effectively detect insurance money laundering pre-payout.
• We Committed Fraud with OpenAI's New Image Model (and Called Mum) - EP99.38 — AI image generation has reached a level of realism enabling highly convincing forgeries, posing significant societal and economic challenges.
• What if your security system could think like a fraudster before they act? — AI systems provide real-time, adaptive fraud detection by learning normal behavior and flagging anomalies.
• How to build Transaction Foundation Models in Banking and Payments — Transaction Foundation Models unify financial intelligence by learning from transaction sequences to generate reusable embeddings.
• Unlocking the relational power of graph neural networks in fraud prevention — Graph Neural Networks enhance fraud detection by modeling relational data, uncovering networked fraud invisible to traditional tabular methods.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by VentureBeat.