MAS to Finalise AI Risk Guidelines Covering Agentic AI

Singapore's approach to AI regulation has been building toward this moment for nine months. In November 2025, the Monetary Authority of Singapore put its proposed Guidelines on AI Risk Management out for consultation; this week, the regulator confirmed the final version is close — and that it will explicitly cover agentic AI, the class of systems that don't just answer questions but take actions. For a financial hub whose banks are already deploying AI agents to onboard wealth clients and support corporate banking, that distinction matters: the rules being finalised now are the ones that will govern software making consequential decisions with a bank's authority.

The guidelines set expectations across three layers: board and senior management oversight, risk management frameworks and processes, and controls spanning the full AI lifecycle. MAS Chairman Gan Kim Yong described the approach as principles-based, letting institutions apply risk management proportionately to how they actually use AI — a stance consistent with how MAS has handled earlier waves of fintech regulation. He was responding to questions about the near-term risks of increasingly autonomous agents, including whether the Safeguards for Agentic Finance at Runtime (SAFR) framework could become a mandatory supervisory requirement.

MAS stopped short of committing to that. The regulator said it will keep developing good practices and toolkits with the industry — through the Future of Finance Institute — while reviewing its supervisory expectations. The hesitation is telling: SAFR, a framework for how agent actions get authorised, when human oversight kicks in, and what gets recorded for consequential decisions, is being treated as a living industry standard rather than a fixed rulebook. The same philosophy produced the AI Risk Management Toolkit, built under Project MindForge with 24 financial institutions, which pairs an Operationalisation Handbook — structured around scope and oversight, AI risk management, AI lifecycle management, and capabilities — with case studies from banks that have operationalised these controls in practice.

The timing is no accident. DBS and OCBC have both shipped agentic AI to customers in recent weeks, and industry surveys keep showing most Singapore firms are unprepared for the accountability questions agents raise. By finalising guidance that covers agentic AI while keeping SAFR voluntary for now, MAS is trying to stay ahead of deployment without strangling it — the same regulatory posture Singapore has used to position itself as a trusted testbed for new financial technology.

Why it matters for Singapore: These guidelines will be the reference point for every bank, insurer and fintech in the city-state as agentic AI moves from pilots to production. The principles-based structure gives MAS room to tighten or loosen expectations as the technology matures, and the industry-built toolkit means much of the implementation burden is shared. Watch for three things in the coming months: the final guidelines' exact language on agent oversight, whether SAFR graduates from voluntary framework to supervisory requirement, and how the BuildFin.ai workgroup turns all of this into templates other institutions can adopt.