AI cyber attacks rise 89 percent as supervisors add runtime controls
The Monetary Authority of Singapore detailed emerging risks from agentic artificial intelligence and cyber threats at the Global FinTech Festival. Regulators highlighted a sixfold rise in high-severity vulnerabilities alongside new frameworks for cross-border payments.
Runtime controls for agentic systems
High-severity software vulnerabilities reached 2,200 this year, a sixfold increase against the prior three-year average, while AI-enabled cyber attacks climbed 89 percent.
In response to autonomous systems taking on consequential operations, MAS introduced the SAFR framework to enforce runtime identity verification, pre-execution control evaluation and comprehensive audit logs.
The central bank also issued supervisory AI Risk Management Guidelines alongside two sectoral handbooks.
To prevent market concentration among dominant lenders, the Pathfin.ai matching platform expanded to over 300 participants, while cross-bank data models are being deployed to detect financial scams in near-real time.
From bilateral links to multilateral networks
Cross-border payment volumes across the bilateral UPI-PayNow linkage between Singapore and India have more than doubled each year since its 2023 launch.
Both jurisdictions are expanding connectivity through Project Nexus to move from standalone bilateral corridors to a multilateral settlement framework.
This network evolution builds on formal cooperation agreements signed between 2018 and 2025, linking MAS supervisory standards with the Reserve Bank of India's FREE-AI framework.
Runtime execution exceeds supervisory reach
Supervisory handbooks provide little defense against sub-second algorithmic exploitation.
Multilateral payment integration accelerates settlement speed but widens cross-border contagion channels.
Central banks are chasing technological velocity with static governance frameworks that cannot enforce real-time containment.