This week’s stories reveal AI moving from pilots to proprietary infrastructure: Revolut has built its own in-house AI research unit and model trained on 80 million customers, while Experian has gone live with an 80-model fraud stack aimed squarely at UK banks. Regulators globally are hardening their stance — ASIC’s new AI corporate plan signals that customer-facing AI will now face conduct scrutiny, not just governance review. Across the board, firms that still treat AI as experimental are increasingly exposed.
Top story: Revolut launches proprietary AI research unit and PRAGMA model, trained on 80 million customer transactions in partnership with Nvidia.
Revolut Builds In-House AI Lab, Debuts PRAGMA Model With Nvidia
FinTech Futures · Generative AI
Revolut has launched a dedicated AI research unit and unveiled PRAGMA, a proprietary model built with Nvidia that decodes financial trends, assesses real-time risk, and delivers personalised product recommendations — trained on transaction data from 80 million customers across 40+ markets. The unit also powers security models that review nearly one billion transactions monthly to combat fraud. For the industry, this signals a shift from buying off-the-shelf AI tools to building sovereign AI infrastructure — a move only a handful of fintechs have the scale to make.
Experian Goes Live With 80-Model AI Fraud Stack for UK Banks
Experian PLC Newsroom · Risk
Experian has launched ‘Transaction Forensics’ in the UK — an AI-powered fraud and AML solution built with Resistant AI that deploys more than 80 AI models to give banks a granular, real-time view of fraud risk across bank-to-bank payments. The product combines Experian’s proprietary consumer and commercial data with Resistant AI’s behavioural analytics, targeting the growing gap between increasingly sophisticated AI-powered fraud attacks and legacy detection systems. It is the first joint product from Experian’s strategic investment in Resistant AI made in July 2025, with further products due across 2026.