A synthetic identity slips through onboarding. A deepfake voice authorizes a wire. A mule network moves money faster than your rules engine can flag it. This isn't hypothetical: financial crime penalties jumped 417% in the first half of 2025, and deepfake-based identity fraud is growing over 2,000% year over year. The problem isn't weaker controls — it's that only 20% of financial services firms say their infrastructure is ready for AI agents at scale, while the fraud they're fighting already is.
Databricks and AWS work with fraud and AML leaders at firms such as NatWest, Nubank, and Eurobank to close that gap. Their answer: start thinking about how to unify your data and AI to enable detection and investigation that can retrain as quickly as fraud evolves.
What that looks like in practice:
- Real-time monitoring on streaming data — transactions scored as they happen
- Graph analytics exposing mule networks' static rules can't see
- Entity resolution catching synthetic identities across fragmented systems
- GenAI investigation agents drafting case narratives and filings — cutting investigation time by a third, with a full audit trail
We'll cover the AWS + Databricks architecture behind this, plus customer results. You'll leave knowing:
- Why rising penalties are a detection-architecture problem, not a "try harder" problem
- The re-architecture pattern leading banks are using — one platform, not stitched-together point tools
- How to keep pace with AI-driven fraud without losing the audit trail examiners expect