Singapore Updates Its Agentic AI Governance Framework as Stablecoin Regulation Moves Toward Legislation
Singapore has updated its Model AI Governance Framework for Agentic AI, adding practical controls for multi-agent systems, third-party agents and automation bias as businesses move from generative AI towards systems capable of acting autonomously.
The Infocomm Media Development Authority (IMDA) published the original framework on 22 January 2026 and updated it on 20 May 2026 after receiving feedback from more than 60 organisations. The revised version also incorporates more than ten real-world deployment case studies.
For fintech and financial-services businesses, the development is particularly relevant because agentic AI can move beyond generating information. IMDA expressly identifies actions such as updating customer databases or making payments as examples of what sufficiently autonomous agents may be able to do.
At the same time, Singapore continues to develop a separate prudential framework for single-currency stablecoins. The Monetary Authority of Singapore (MAS) finalised the policy design in 2023, including reserve, capital, redemption and disclosure standards. However, businesses should distinguish that finalised policy framework from the legislation required to make the dedicated stablecoin issuance regime operational.
The result is a useful illustration of Singapore’s wider regulatory model: emerging technologies are not necessarily governed through the same legal instrument. AI agents are currently addressed through practical governance guidance, while stablecoin issuance is moving towards a more conventional financial-services licensing framework.
For readers’ convenience, we have placed the key official AI and stablecoin regulatory materials at the end of this article.
Publish Date
29 Aug 2026
Reading Time
11 minutes
Category
Legal News
Jurisdiction
Singapore
Singapore has expanded its governance model for agentic AI
The Model AI Governance Framework for Agentic AI builds on Singapore’s earlier AI governance work but addresses a materially different technological capability.
Traditional generative AI largely responds to prompts. Agentic systems can plan across several steps, use tools and take actions to pursue specified objectives. IMDA defines agentic AI systems as systems capable of planning across multiple steps using AI agents.
The framework first launched in January 2026
IMDA launched the framework at the World Economic Forum on 22 January 2026.
It is intended for organisations developing their own AI agents as well as businesses deploying third-party agentic solutions. The framework focuses on technical and organisational controls rather than creating an AI licensing regime.
IMDA organises the framework around four broad dimensions:
- assessing and bounding risks before deployment;
- maintaining meaningful human accountability;
- implementing technical controls throughout the agent lifecycle; and
- enabling responsible end-user use through transparency and training.
The May update makes the framework more operational
On 20 May 2026, IMDA issued an updated version reflecting feedback from more than 60 organisations, including technology companies, financial institutions and other enterprises.
The update adds guidance on multi-agent systems, third-party agent risks and automation bias. It also includes more than ten real-world case studies showing how organisations have translated the framework into operational controls.
IMDA’s current AI portal identifies the updated framework as version 1.5.
That evolution matters because agentic AI governance is moving quickly from high-level principles towards practical questions such as who may approve an action, what an agent can access, how permissions are limited and how autonomous activity is monitored.
Agentic AI creates a different type of operational risk
The regulatory interest in agentic AI comes from its ability to affect the external environment.
An AI system that only drafts text creates one risk profile. An agent that can access a payment system, modify customer data, interact with third-party APIs or execute instructions creates a substantially different one.
Financial actions raise the stakes
IMDA specifically notes that AI agents may have access to sensitive data and may be capable of actions such as updating customer databases or making payments.
The corresponding risks include erroneous or unauthorised actions and an increased possibility that users over-trust systems that have previously performed reliably.
For financial firms, this creates a direct link between AI governance and existing control functions.
An AI agent involved in payment initiation, treasury operations, customer onboarding or transaction monitoring does not operate outside financial regulation simply because the underlying action is automated.
Autonomy should be bounded by risk
One of IMDA’s core recommendations is to assess suitable agentic use cases in advance and restrict the agent’s powers accordingly.
This can include limits on:
- access to data;
- access to tools and external services;
- permitted actions;
- degree of autonomy; and
- situations in which human approval is required.
The approach is deliberately risk-based rather than binary.
Not every automated action needs identical approval. A reversible low-impact task may justify materially more autonomy than a transaction capable of moving customer funds or changing sensitive access permissions.
Human accountability remains central
Singapore’s framework does not treat greater AI autonomy as a reason to remove organisational accountability.
Instead, IMDA emphasises that humans remain ultimately accountable for agent deployment.
Significant checkpoints should be identified
Businesses should determine where human intervention is necessary rather than inserting approvals arbitrarily into every step.
The May update provides practical examples. In the Dayos case study, different actions are assigned different risk tiers according to factors such as impact, reversibility and feasibility of human oversight.
Low-risk actions can be automated, while more significant actions require approval or may be prohibited entirely.
That model can translate naturally into financial-services controls.
An internal agent may be permitted to retrieve information automatically, while changing payment instructions, releasing funds or altering customer permissions may justify materially stronger authorisation.
Human approval also needs to be meaningful
Simply inserting a human click into an automated process does not necessarily create effective oversight.
The updated framework highlights the importance of giving human reviewers enough information to understand what the agent intends to do and why approval is being requested.
For regulated firms, this should align closely with existing concepts such as maker-checker controls, segregation of duties and delegated authority.
Third-party and multi-agent systems create additional governance challenges
One of the most useful additions in the May update concerns systems that do not operate as a single internally controlled agent.
Modern enterprise deployments may combine proprietary models, third-party agents, external APIs and multiple autonomous systems.
Third-party agents do not remove the organisation’s responsibility
A business may not control every component of a third-party agent.
It can nevertheless control decisions such as:
- which systems the agent may access;
- what information it can receive;
- which external services it can call;
- what actions require human approval; and
- how its activity is logged and monitored.
IMDA’s updated framework expressly includes third-party agent risks among the areas requiring additional governance attention.
For financial institutions, this is especially relevant where agents connect to cloud infrastructure, KYC providers, payments APIs, customer databases or other regulated technology dependencies.
Multi-agent systems complicate accountability
Multi-agent environments can also make it harder to determine which component initiated an action or produced a decision.
IMDA’s May update therefore expands the framework to address risks arising from interactions among multiple agents.
Businesses adopting such systems should consider whether audit logs, identity controls and system architecture make it possible to reconstruct what occurred after an incident.
Singapore’s agentic AI framework is not an AI licensing regime
The legal status of the framework is important.
The Model AI Governance Framework for Agentic AI is official governance guidance. It is not equivalent to a statute, mandatory licensing framework or Singapore version of the EU AI Act.
IMDA itself describes Singapore’s wider approach as practical and risk-based rather than top-down, with governance frameworks and assurance tools developed alongside industry.
Existing laws still apply to AI-enabled activities
Non-statutory AI guidance should not be interpreted as meaning AI deployment is legally unrestricted.
Existing obligations may continue to apply depending on the activity, including requirements relating to financial services, payments, data protection, cybersecurity, consumer protection and contractual responsibility.
The practical consequence is that organisations need two separate analyses:
How should the AI agent be governed?
and
What legal rules apply to the activity the agent is performing?
For fintech firms, the second question can be just as important as the first.
Singapore’s stablecoin framework sits at a different regulatory stage
The position for stablecoins is structurally different.
MAS finalised its single-currency stablecoin policy framework on 15 August 2023 following consultation.
The framework is intended to apply to single-currency stablecoins pegged to the Singapore dollar or a G10 currency and issued in Singapore.
MAS designed four core regulatory pillars
The published framework focuses on:
- value stability;
- capital and prudential requirements;
- redemption at par; and
- disclosure.
These requirements are intended to create a differentiated category of stablecoin capable of being recognised as an “MAS-regulated stablecoin.”
However, the framework’s detailed prudential model should be distinguished from the existing regulation of digital payment token activities under the Payment Services Act 2019.
What MAS expects from regulated single-currency stablecoins
Although legislative implementation remains important, the policy design provides businesses with a relatively clear picture of the standards MAS intends to apply.
Reserves must fully support circulating stablecoins
MAS states that reserve assets should be held in low-risk and highly liquid assets.
Their value must be at least 100% of the value of the single-currency stablecoins in circulation at all times, and the assets should be held in segregated accounts with eligible custodians.
This is designed to support confidence that the issuer can meet redemption obligations rather than relying on an unstable or opaque backing mechanism.
Prudential requirements include minimum capital
The published framework specifies minimum base capital of S$1 million, together with minimum liquid assets intended to cover operating expenses.
MAS also contemplated restricting regulated issuers from conducting unrelated non-issuance business that could introduce additional risks into the issuing entity.
The regulatory logic is straightforward: stablecoin holders should not be indirectly exposed to unnecessary risks arising from unrelated business lines.
Redemption must be available at par
MAS intends regulated SCS holders to have a direct legal claim for redemption at par.
The published standard requires timely redemption no later than five business days after a redemption request, subject to reasonable conditions disclosed in advance.
This is one of the clearest differences between a regulated payment instrument and a token that merely attempts to maintain a market price close to a fiat currency.
Disclosure supports the value-stability model
The framework also expects an online white paper explaining key matters such as the stablecoin’s value-stabilisation mechanism and holders’ rights.
Reserve information and audit-related disclosures form part of the broader transparency model described by MAS.
The “MAS-regulated stablecoin” label has a specific regulatory purpose
MAS intends the label “MAS-regulated stablecoin” to allow users to distinguish qualifying SCS from other digital payment tokens marketed as stablecoins.
The 2023 framework states that only issuers satisfying the applicable requirements would be able to apply for recognition under that label.
This distinction is commercially important.
A stablecoin may be available in Singapore without necessarily being an “MAS-regulated stablecoin” under the dedicated SCS framework.
Businesses should therefore avoid treating the words “stablecoin”, “regulated digital asset” and “MAS-regulated stablecoin” as interchangeable.
The dedicated SCS framework should not yet be described as fully operative law
This is the most important legal-status distinction.
MAS finalised the policy features of the framework in 2023. But in November 2025, MAS Managing Director Chia Der Jiun stated that MAS had finalised the features of the regime and would be preparing draft legislation.
Current 2026 regulatory analysis continues to describe legislation implementing the dedicated stablecoin issuance service as being finalised rather than already fully operative.
As of our 26 August 2026 freshness check, we have not identified an official MAS announcement or Singapore commencement instrument confirming that the dedicated SCS issuance regime described in the 2023 policy framework has become fully operative.
Accordingly, businesses should not base a licensing analysis on the assumption that every feature of the finalised policy document already constitutes binding law.
Existing payments and digital-token regulation still applies
The absence of confirmed commencement of the dedicated SCS regime does not mean stablecoin-related activity is outside Singapore financial regulation.
MAS already regulates a broad range of digital payment token services under the Payment Services Act 2019.
From 4 April 2024, amendments expanded the regulated DPT perimeter to include activities such as:
- DPT custody;
- facilitation of DPT transfers;
- facilitation of DPT exchange even without taking possession of the assets; and
- certain cross-border money-transfer activities.
MAS can impose AML/CFT, user-protection and financial-stability requirements on relevant DPT service providers.
For a stablecoin business, the practical analysis is therefore not simply “Has the future SCS regime commenced?”
It must also assess whether existing activities already fall within the current payments or digital-token licensing perimeter.
Agentic AI and stablecoins illustrate two different regulatory strategies
| Area | Regulatory approach | Current position |
|---|---|---|
| Agentic AI | Official model governance framework | Non-statutory risk-based guidance, updated 20 May 2026 |
| Dedicated SCS issuance framework | Prudential and licensing model | Policy design finalised; formal legislative implementation must be distinguished from the policy framework |
| Existing DPT/payment services | Payment Services Act 2019 and related rules | Existing binding regulatory perimeter |
A firm can voluntarily adopt stronger agentic AI governance controls without obtaining an “AI licence”. By contrast, conducting a regulated payment or digital-token service can require statutory permission even when AI is merely the technology used to provide it.
Where agentic AI and digital finance intersect
The two subjects become particularly relevant when an AI agent is permitted to interact with financial infrastructure.
Agents initiating or preparing payments
An agent may potentially:
- prepare payment instructions;
- select payment routes;
- reconcile transactions;
- initiate transfers; or
- interact with treasury systems.
IMDA’s risk-bounding approach suggests that organisations should decide in advance which of those actions can occur autonomously and which require meaningful human approval.
Financial-regulatory requirements remain separate.
If the underlying activity falls within a regulated payments function, delegating part of the process to an agent does not transfer regulatory accountability from the business to the software.
Agents interacting with digital assets or stablecoins
Similar issues arise where agents interact with wallets, digital assets or tokenised settlement infrastructure.
Potential use cases can include treasury management, settlement, reconciliation, customer servicing and compliance workflows.
Businesses should consider:
- what wallet or account permissions an agent receives;
- whether transaction limits are imposed;
- which instructions require human confirmation;
- what third-party systems are accessible;
- whether actions are logged in sufficient detail; and
- whether an agent can override or bypass existing controls.
The relevant technical architecture should align with the firm’s regulatory permissions and operational-risk framework.
What fintech businesses should assess now
Businesses deploying agentic AI
Financial institutions and fintech companies should map an agent’s actual powers rather than treating “AI use” as a single risk category.
Particular attention should be given to data access, system permissions, payment authority, irreversible actions, third-party integrations and human checkpoints.
Prospective stablecoin issuers
Businesses considering Singapore issuance can already prepare against the policy architecture published by MAS.
Reserve structures, capital planning, redemption processes, disclosure and operational separation can all be incorporated into early feasibility analysis.
However, licensing strategy should be refreshed against the actual legislation and commencement status at the point of application rather than relying solely on the 2023 policy framework.
Existing crypto and DPT businesses
Businesses should also test their activities against the current Payment Services Act perimeter.
Stablecoin custody, transfer, dealing or exchange activity may raise existing licensing issues independently of the future dedicated SCS issuance framework.
Cross-border groups
International fintech groups should avoid assuming that one Singapore entity or licence automatically covers AI deployment, token issuance, custody, dealing and payment services.
Each activity has its own regulatory analysis.
Legasset advises fintech, payments and digital-asset businesses on regulatory structuring, licensing strategy, AML/CFT, stablecoin perimeter analysis and market entry. We can also support businesses assessing governance and regulatory controls where AI agents are introduced into regulated financial workflows.
Singapore Agentic AI and Stablecoin Regulation FAQ
Is Singapore’s Agentic AI Governance Framework legally binding?
The Model AI Governance Framework for Agentic AI is an official governance framework issued by IMDA, not a standalone licensing statute.
It provides recommended technical and organisational measures for responsible agent deployment while emphasising human accountability.
What changed in the May 2026 agentic AI framework update?
The 20 May 2026 update incorporated feedback from more than 60 organisations and added new guidance on multi-agent systems, third-party agents and automation bias.
It also added more than ten real-world deployment case studies showing how organisations have implemented different parts of the framework.
Can an AI agent be allowed to initiate payments?
IMDA does not impose a universal prohibition on agent-initiated payments.
Instead, it identifies payment activity as an example of an action that can create heightened risk and recommends bounding autonomy, restricting access and using meaningful human approval at significant checkpoints.
Any existing financial-services rules applicable to the payment activity continue to apply separately.
What are the four core areas of Singapore’s agentic AI framework?
The framework focuses on risk assessment and boundaries, meaningful human accountability, technical lifecycle controls, and end-user transparency and education.
Is Singapore’s dedicated stablecoin framework already fully in force?
The policy framework has been finalised, but the legal status must be described carefully.
MAS stated in November 2025 that it had finalised the features of the stablecoin regime and would prepare draft legislation. During our 26 August 2026 freshness check, we did not identify an official commencement announcement confirming that the dedicated SCS issuance regime had become fully operative.
Which stablecoins does the MAS policy framework cover?
The framework focuses on single-currency stablecoins pegged to the Singapore dollar or a G10 currency and issued in Singapore.
What reserve backing does MAS expect?
The policy framework requires reserve assets to be low risk and highly liquid, with a value of at least 100% of SCS in circulation at all times.
The reserves should also be segregated and held with eligible custodians.
What is the minimum capital requirement under the stablecoin framework?
MAS’s published policy framework specifies S$1 million in minimum base capital, together with liquid-asset requirements.
The applicable legal requirement at the point of licensing should be checked against the final implementing legislation once operative
How quickly must an MAS-regulated stablecoin be redeemable?
The published framework provides for redemption at par within five business days of a valid redemption request.
Are other stablecoins unregulated in Singapore?
No.
Stablecoins are digital payment tokens, and activities involving them can fall within existing regulation under the Payment Services Act 2019 depending on the service being provided. The dedicated SCS framework adds a specialised value-stability and issuer regime rather than creating Singapore’s entire digital-token regulatory perimeter from scratch.
Singapore Digital Governance: Official AI and Stablecoin Regulatory Materials
The 22 January 2026 announcement explains why Singapore introduced dedicated guidance for autonomous AI agents and sets out the framework’s four core governance dimensions.
II. Infocomm Media Development Authority — Updated Model AI Governance Framework for Agentic AI
The 20 May 2026 update adds guidance on multi-agent systems, third-party agents and automation bias, together with more than ten real-world deployment case studies.
III. Infocomm Media Development Authority — Artificial Intelligence in Singapore
IMDA’s current AI portal provides access to the latest version of the agentic AI framework and Singapore’s wider AI governance and assurance resources.
IV. Monetary Authority of Singapore — MAS Finalises Stablecoin Regulatory Framework
The official MAS announcement sets out the scope of the single-currency stablecoin policy framework and its key requirements on value stability, capital, redemption and disclosure.
V. Monetary Authority of Singapore — Stablecoin Regulatory Framework Requirements
The official framework summary confirms 100% reserve backing, the S$1 million minimum base-capital requirement, redemption within five business days and core disclosure expectations.
VI. Monetary Authority of Singapore — Expanded Regulation of Digital Payment Token Services
This MAS announcement explains the existing Payment Services Act perimeter for digital-token custody, transfer, exchange and related services, which remains relevant independently of the dedicated SCS framework.
VII. Monetary Authority of Singapore — Creating the Future of Finance: Stablecoins and Tokenised Money
In these November 2025 remarks, MAS Managing Director Chia Der Jiun states that MAS had finalised the features of the stablecoin regime and would be preparing draft legislation, providing important context on its implementation status.
