MAS Taps Shared Bank Data to Train Scam-Detecting AI
Source: Fintech News SG
MAS is building AI that learns from data pooled across banks and the public sector to flag suspect accounts and transactions sooner, with findings due by end-2026.
Singapore's central bank is training AI models on data pooled across banks and the public sector to flag suspect accounts and transactions sooner. The Monetary Authority of Singapore is working with law enforcement and lenders on the push, with findings expected by the end of 2026.
Managing director Chia Der Jiun laid out the effort at Global Fintech Fest in India, calling its aim earlier detection, faster intervention and smaller losses. Local lenders already apply machine learning to fraud screening, credit decisions, risk oversight, compliance and customer service, several at scale.
But Chia said the technology still falters where work needs nuanced reading, commercial judgement or human contact, and only some adopters have booked big productivity gains. He expects that to change as staff train more and organisations rebuild processes and products around AI. He also wants to keep advanced tooling from concentrating among the largest players, citing Pathfin.ai, a MAS programme that helps smaller firms find vetted AI products and now counts over 300 participants.
On governance, MAS published its generative AI risk framework back in 2023, then two AI Risk Management Handbooks during 2025, and has drafted AI Risk Management Guidelines for consultation on lifecycle governance, risk and controls. A July 2026 SAFR white paper covers autonomous agents: they must confirm identity and authority, have planned actions vetted before running and leave an audit trail. Chia also warned that stronger models help attackers, noting high-severity Common Vulnerabilities and Exposures climbed sixfold to 2,200 in 2026 against the prior three-year average, while CrowdStrike logged an 89% rise in AI-enabled attacks - though breaches have not grown as quickly, a gap he credits to model safeguards and layered defences.
AI is outpacing other technologies on Chia's radar: tokenisation needs more years to scale, quantum may be five to ten years out, and he says resilience work should start now. Momentum is clearest on the Singapore-India corridor, where the 2023 PayNow-UPI link doubles transaction volumes yearly and both nations back Nexus, a framework linking instant payment systems. Pints AI and a large Indian insurer used AI to speed checks and brief underwriters, leaving calls to experts, and are drafting a sector white paper.
Why it matters for Singapore: The city-state is staking its fintech edge on writing AI rules while keeping smaller players in the game. If pooled bank data sharpens scam detection, consumers and lenders both gain, and MAS's handbook-led approach hands other regulators a template to follow.
