Federated Learning & Differential Privacy: Architects for Secure AI Collaboration
Wednesday, July 15th, 2026: 10:00 AM to 11:00 AM
In this session, we'll cover the architectural patterns making secure, multi-party AI collaboration a production reality. We'll walk through how federated learning keeps raw data secure inside each institution's boundaries, how differential privacy provides provable guarantees against re-identification, and how these techniques work together with accelerated computing to power real-time AML, fraud, and sanctions workflows.
Virtual
Financial institutions hold some of the richest transaction, customer, and SAR data in the world—as well as the strictest constraints on using it. The result is a structural blind spot in financial crime detection.
The UN estimates less than 1% of laundered funds are ever seized, while compliance teams drown in false positives from models that only see one institution's slice of activity.
FinCEN's 2026 proposed AML/CFT reforms now explicitly reward institutions that use AI to demonstrate program effectiveness, and regulators globally are sharpening expectations around explainability, bias, and data privacy.
What You’ll Learn:
- Why siloed AML models miss layered laundering patterns
- How to map federated learning and differential privacy onto existing model risk management controls (SR 11-7)
- What a reference architecture for federated financial crime detection looks like in production, from data residency to audit trail
- How these approaches align with FinCEN's effectiveness framework, the EU AI Act, and FCA explainability expectations
Hosted by Informa