← Back to Home News · SG

MAS keeps agentic AI under its proposed AI guidelines, not binding rules

Make RegTech.com preferred on Google
MAS keeps agentic AI under its proposed AI guidelines, not binding rules
In a 5 August 2026 parliamentary reply, MAS said its proposed AI Risk Management Guidelines cover all AI uses by financial institutions, including agentic AI.
AI Summary
  • In a written reply to Parliament on 5 August 2026, MAS confirmed its proposed Guidelines on AI Risk Management will apply to all AI use cases by financial institutions, including agentic AI.
  • The Guidelines, issued for consultation in November 2025, are still to be finalised and set supervisory expectations rather than legally binding rules.
  • Asked whether it would move to mandatory requirements, MAS reaffirmed its principles-based approach and did not commit to binding obligations.
  • Industry-led work, Project MindForge's AI Risk Management Toolkit and the SAFR framework, supports practical safeguards for how AI agents are authorised.

When a Member of Parliament asked whether Singapore intended to write hard rules for the autonomous AI agents starting to appear in financial services, the Monetary Authority of Singapore gave a deliberately measured answer: the technology is already caught by the risk guidance it is preparing, and for now that guidance sets expectations rather than binding obligations.

In a written reply to Parliament on 5 August 2026, MAS, answering through its Chairman, Deputy Prime Minister Gan Kim Yong, confirmed that its proposed Guidelines on AI Risk Management “apply to all AI use cases by FIs, including agentic AI.” Those Guidelines were issued for public consultation in November 2025 and, as the reply notes, are still to be finalised.

What was asked, and what MAS said

MP Mariam Jaafar's question had three parts: how MAS reads the near-term risks from increasingly autonomous AI agents, whether the regulator would move from the current industry-led approach to mandatory supervisory requirements, and, if so, on what timeline.

MAS’s reply leaned on continuity rather than a new departure. It described a principles-based approach to guiding safe and responsible AI adoption, and said the proposed Guidelines set out supervisory expectations for board and senior management oversight, sound risk management, and controls across the AI life cycle. Crucially, it added that these expectations cover the full range of AI uses, including agentic systems, and would be finalised soon.

On the specific question of making the regime mandatory, MAS did not commit. It reaffirmed its principles-based stance and said it would continue to review its supervisory expectations and update them where necessary. 

Guidelines are expectations, not binding law

This distinction matters, because some of the coverage that followed described the reply as confirmation of binding rules for AI agents. That is not what the instrument is. In the MAS framework, Guidelines set out supervisory expectations, a form of soft law: financial institutions are expected to observe them, and departures invite supervisory scrutiny, but they do not carry the legal force of a Notice or a Regulation. On top of that, the Guidelines here are still at the proposed stage and not yet in effect.

The accurate reading is therefore narrower, and more interesting, than the loudest headlines. Agentic AI sits within the scope of a proposed, non-binding framework, and MAS has chosen, at least for this cycle, to hold the line on expectations rather than reach for a mandate.

The industry-led layer

Alongside the Guidelines, MAS pointed to work carried out with the industry to make the expectations practical. Under Project MindForge, firms have built an AI Risk Management Toolkit to help implement the Guidelines. A separate piece of work, the Safeguards for Agentic Finance at Runtime framework, sets out a possible way to handle how an AI agent’s actions are authorised when they are taken. Both are industry-developed and operational, sitting beneath supervisory expectations rather than replacing them.

Why it matters

The reply is a useful read on how one of the more closely watched financial regulators is choosing to handle agentic AI. Rather than rushing a bespoke rulebook out ahead of a fast-moving technology, MAS is extending an existing principles-based framework to cover it and keeping pressure on firms through supervisory expectations, while leaving the door open to firmer rules later. For financial institutions, the near-term task is concrete: prepare for the finalised Guidelines on governance, risk management and life-cycle controls, applied to agentic systems, rather than wait for a new set of statutory obligations.

It also tempers the more sweeping claims made about the reply. MAS did not announce the first binding regime for autonomous agents; it confirmed that its proposed guidance already reaches them, and it kept the option of mandatory requirements in reserve. The Guidelines are still to be finalised, and MAS said it will keep reviewing and updating its expectations as the technology develops.

Link copied to clipboard
AI Bot
AI Bot
Hi! I'm the RegTech.com assistant. How can I help you today?
Ask me anything — top trending news, latest regulatory changes in the EU, or simply search for topics.