Revolut launches AI research unit for banking models

Yesterday, Revolut announced PRAGMA — a proprietary foundation model trained on 80 million customer records across 40 markets, built with NVIDIA to unify fraud detection, risk assessment, and customer service.

Here's why this matters: Revolut Research didn't customize a third-party LLM. They built domain-specific infrastructure from architecture up.

That distinction separates production AI from pilot theater.

The compliance question nobody's asking:

When your fraud detection, risk engine, and customer service share a single foundation model, how do you prove data isolation to regulators?

APRA CPS 230 requires you to demonstrate operational resilience at the architecture level. A unified model creates elegant engineering — and a single point of regulatory scrutiny.

The data from 80 million customers becomes your moat. It also becomes your attestation burden.

What regulated enterprises should watch:

→ How Revolut structures model governance across three distinct risk domains on shared infrastructure
→ Whether their audit trail separates training provenance from inference decisions
→ How they demonstrate compliance boundaries inside a unified system

At QikAI, we work with financial services teams navigating exactly this trade-off: domain-specific models deliver better outcomes, but only if your governance architecture anticipates the regulator's second question.

Build vs. buy isn't the hard part anymore. Build and prove it to your board is.

What's your take — does a unified model make compliance easier or harder?

#AIGovernance #FinancialServices #RegulatoryCompliance #EnterpriseAI #CPS230

📩 The full weekly breakdown lands in AI in Telecom, Banking & Retail → https://amplyfy.app/wire/subscribe/18