MAS Moves Financial AI Governance From Model Principles to Runtime Controls

MAS Managing Director Chia Der Jiun set out a financial-system agenda built around trust, connectivity and resilience, with artificial intelligence identified as the most immediate technology priority. The address connected industry adoption to AI risk guidelines, shared solution discovery, runtime controls for agentic finance and system-level fraud detection.

The speech was delivered at the Global FinTech Fest on 11 September 2026. MAS positioned Pathfin.ai as an industry matching platform for validated AI solutions, referred to earlier generative AI risk work and AI Risk Management Handbooks, and described proposed supervisory expectations across governance, risk management and the AI life cycle.

The institutional shift is from broad statements of responsible AI toward controls that can be tested during deployment. Financial institutions need evidence that an AI system has an authorised purpose, operates within defined limits, produces reviewable records and can be challenged when outputs affect customers, markets or regulatory obligations.

Banks and market operators should map each use case to accountable owners, approved data, model and agent permissions, pre-execution checks, exception handling, monitoring and shutdown conditions. Participation in shared initiatives such as Pathfin.ai or the Safeguards for Agentic Finance at Runtime work should be translated into institution-specific controls rather than treated as external assurance.

MAS reported more than 300 participants on Pathfin.ai and said successful matches were increasing. It also described work with law enforcement and the banking industry to test AI against cross-bank and public-private data for near-real-time detection of suspicious accounts and transactions, with findings expected by the end of 2026. These are programme indicators, not proof of system-wide effectiveness.

Singapore is using regulatory coordination and common infrastructure to lower the cost of trustworthy AI adoption across institutions. For ASEAN, the model is relevant where cross-border payments, fraud networks and differing levels of institutional capability make isolated controls insufficient. Interoperability will depend on common identity, data-use and audit expectations.

Watch for the final MAS AI risk guidance, results from the cross-bank fraud experiments, measurable Pathfin.ai adoption outcomes and further specifications under the SAFR initiative. The key test is whether supervisory expectations become comparable evidence across different models, vendors and institutions.