MAS tests cross-bank AI models to detect suspicious accounts; results due by year-end

Facts

MAS managing director Chia Der Jiun spoke at Global Fintech Fest 2026 in Mumbai on 11 September 2026 (transcript republished by the BIS on 21 September). According to the transcript, MAS is testing several AI models using cross-bank data and public-private data to improve detection of suspicious accounts, with results expected by end-2026 (source: BIS transcript).

He also said Pathfin.ai, a platform for smaller financial institutions to find vetted AI solutions, has more than 300 participants, and that MAS published SAFR, a framework of runtime safeguards for agentic finance, as a white paper in July 2026. On cross-border payments, he said transaction volumes on the PayNow–UPI link with India have more than doubled every year since launch (source: same). ¹

Analysis

Over time, MAS issued AI risk-management guidelines and an industry handbook in 2025, published SAFR in July 2026, and is now piloting detection on pooled data. The focus is moving from governance to jointly operated fraud detection. The underlying problem is that scams using mule accounts (accounts that receive criminal proceeds) are hard to spot with one firm's data. As cross-border instant payments spread, funds can move through multiple accounts and countries within minutes, increasing the need for joint detection.

Implications

Counterpoints and uncertainties

The speech did not say which institutions are in the pilot, the legal basis for data sharing, or whether non-banks and licensed moneylenders can participate. The pilot may not lead to full deployment. A speech signals policy direction and is not a regulatory obligation.