AI & Automation

AI Governance Framework Singapore: The Official Documents, What Is Voluntary and What to Produce

November 8, 2026 · 8 min read · Tim Etzal Group

Abstract illustration of stacked translucent layers with a shield outline resting on the top layer

The AI governance framework Singapore companies refer to is not a single law but a family of official documents: the Model AI Governance Framework (second edition, 21 January 2020), the Model AI Governance Framework for Generative AI (30 May 2024), the Model AI Governance Framework for Agentic AI (first published January 2026), and the AI Verify testing framework and toolkit. All of these are voluntary guidance; the binding layer underneath is the Personal Data Protection Act 2012 (PDPA), plus sector rules such as MAS's guidelines for financial institutions. A mid-size company should be able to show a use-case inventory, a risk rating and level of human oversight per use case, named owners, and PDPA checks for any AI that touches personal data.

Updated: 8 November 2026

What AI governance means

AI governance is the set of roles, rules and records an organisation uses to decide which AI it uses, how risky each use is, who is accountable for it, and how it is checked over time. Singapore's approach, led by the Infocomm Media Development Authority (IMDA) and the Personal Data Protection Commission (PDPC), is to publish practical frameworks and testing tools that organisations adopt voluntarily, alongside existing law such as the PDPA. IMDA's artificial intelligence page lists them in one place.

The AI governance framework Singapore publishes: the core documents

Model AI Governance Framework, second edition (2020)

The first edition was released for consultation on 23 January 2019, and the second edition on 21 January 2020. The PDPC describes it as a "living and voluntary" framework and "a general, ready-to-use tool" for organisations deploying AI at scale (paragraph 2.2 of the second edition). It rests on two principles: decisions made by AI should be explainable, transparent and fair, and AI systems should be human-centric.

It organises the work into four areas: internal governance structures and measures; determining the level of human involvement in AI-augmented decision-making; operations management; and stakeholder interaction and communication. Two details matter for mid-size firms. It is not intended for organisations that simply deploy updated off-the-shelf software that now happens to include AI features. And adopting it "will not absolve organisations from compliance with current laws and regulations" (paragraph 2.12). Its companion, the Implementation and Self-Assessment Guide for Organisations (ISAGO), helps you check how far your practices align.

Model AI Governance Framework for Generative AI (2024)

Developed by the AI Verify Foundation and IMDA and released on 30 May 2024, it extends the earlier framework to generative AI across nine dimensions: accountability, data, trusted development and deployment, incident reporting, testing and assurance, security, content provenance, safety and alignment R&D, and AI for public good. For a company that deploys generative AI tools rather than building models, accountability, data, incident reporting, testing and security are the dimensions that become internal work.

Model AI Governance Framework for Agentic AI (2026)

IMDA first published it in January 2026 and updated it in May 2026 (version 1.5), adding case studies and practices on multi-agent systems, third-party agents and automation bias. It is for organisations deploying AI agents, whether built in-house or bought, and covers four dimensions: assess and bound the risks upfront, make humans meaningfully accountable, implement technical controls and processes, and enable end-user responsibility. Its central point for buyers: humans stay accountable, and human approval belongs at significant checkpoints such as high-stakes or irreversible actions. Our guide to build vs buy AI agents applies this workflow by workflow.

AI Verify: testing framework and toolkit

AI Verify is an AI governance testing framework and software toolkit developed by IMDA. The framework sets out 11 principles, from transparency and explainability to fairness, accountability and human agency and oversight, each with outcomes, process checks and documentary evidence. For traditional machine learning models, the toolkit runs technical tests for fairness, explainability and robustness; Project Moonshot covers generative AI.

According to the AI Verify Foundation, a not-for-profit body set up by IMDA, AI Verify was launched for international pilot on 25 May 2022, open-sourced on 7 June 2023 and updated to cover generative AI on 29 May 2025. It works as a self-assessment: start with the published process checks, mark each one as implemented or not, note where the evidence lives, run the technical tests only where you have a model to test, and produce a report for management, auditors or customers. Nothing on the official pages presents it as a certification or as IMDA approval of your system.

Voluntary or binding: where the law actually sits

The frameworks above are guidance. The obligations come from elsewhere:

  • The PDPA is law. It applies to AI whenever personal data is collected, used or disclosed. PDPC's Advisory Guidelines on use of Personal Data in AI Recommendation and Decision Systems (1 March 2024) explain how; the guidelines are advisory and not legally binding, but the PDPA they interpret is. PDPC added guidelines for generative AI on 20 July 2026.
  • Sector rules apply on top. The 2020 framework notes that adopting it does not mean compliance with sector-specific rules in fields such as finance, healthcare and law (paragraph 2.13). For financial institutions, MAS published Guidelines on Artificial Intelligence Risk Management on 7 October 2026, setting out its supervisory expectations; they take effect on 7 October 2027.

Which document answers which question

DocumentIssued by, dateStatusThe question it answers
Model AI Governance Framework, 2nd editionPDPC and IMDA, 21 January 2020VoluntaryHow do we govern AI decisions inside the company
Framework for Generative AIAI Verify Foundation and IMDA, 30 May 2024Voluntary guidanceWhat changes when the AI generates content
Framework for Agentic AIIMDA, January 2026, v1.5 May 2026Voluntary guidanceHow far can an AI agent act before a person approves
AI Verify testing framework and toolkitIMDA and AI Verify Foundation, updated 29 May 2025Voluntary self-assessmentCan we show evidence the system was tested
Personal Data Protection Act 2012Law, administered by PDPCBindingCan we use this personal data, and what must we tell people
PDPC AI guidelines (2024) and GenAI guidelines (2026)PDPC, 1 March 2024 and 20 July 2026Advisory, interpreting binding lawHow the PDPA applies to training, deploying and buying AI
MAS Guidelines on AI Risk ManagementMAS, 7 October 2026Supervisory expectations for financial institutionsWhat MAS expects a financial institution to have in place

What a mid-size company should actually produce

The 2020 framework says the extent of adoption depends on the nature and complexity of the AI, how far it is used in decisions, and the severity and probability of the impact on individuals (paragraph 2.5). In practice, that means a short set of documents:

  1. AI use-case inventory: every AI tool or feature in use or planned, its owner, its data and its vendor.
  2. Risk rating per use case: probability and severity of harm, plus access, reversibility and autonomy when the AI takes actions.
  3. Level of human involvement per use case: human-in-the-loop, human-over-the-loop or human-out-of-the-loop, with the reason.
  4. Accountability map: who approves new AI uses, who owns each one, who handles an incident.
  5. Staff AI use policy: approved tools, which data may go into prompts, what must be checked before output leaves the company.
  6. PDPA record: for each use case touching personal data, the consent or exception relied on and the notification wording.
  7. Vendor file: what each AI vendor stores, where and for how long, and the safeguards it can show.
  8. Test and review log: what was tested before go-live, what is monitored, and the next review date.

Still deciding where AI is worth applying? What an AI readiness assessment should deliver covers that earlier step.

Key takeaways

  • Singapore's AI governance framework is a family: the 2020 Model Framework, the 2024 generative AI framework, the 2026 agentic AI framework and AI Verify.
  • All of them are voluntary. The PDPA is binding, and MAS sets supervisory expectations for financial institutions.
  • Every framework asks the same question: how risky is this use case, and how much human involvement does it need.
  • A mid-size company needs an inventory, a risk rating and oversight level per use case, named owners, a staff policy, PDPA records and a vendor file.
  • AI Verify is a self-assessment and testing tool, not a certification.

AI governance with Etzal Group

Etzal Group's AI Consulting package can produce the artefacts above: the use-case inventory, risk tiering, level of human involvement and accountability map, aligned to the IMDA and PDPC frameworks. The frameworks are voluntary, and the work does not make your company certified, IMDA-endorsed or a member of the AI Verify Foundation. We are not a law firm, so the legal reading of the PDPA for a specific case stays with your data protection officer or counsel. Our questions to ask before hiring an AI consultant apply to us as well. A first conversation is free: book a consultation.

Frequently asked questions

What is Singapore's AI governance framework?

It is a family of official, voluntary documents: the Model AI Governance Framework (second edition, 2020), the Model AI Governance Framework for Generative AI (2024), the Model AI Governance Framework for Agentic AI (2026) and the AI Verify testing framework. The binding layer underneath is the PDPA, plus sector rules such as MAS guidelines for financial institutions.

Is Singapore's AI governance framework mandatory?

No. The PDPC calls the Model Framework "living and voluntary", and the generative and agentic AI frameworks are published as guidance. The PDPA is mandatory, though, and applies whenever an AI system uses personal data.

What is Singapore's Model AI Governance Framework?

It is guidance from IMDA and the PDPC, first released on 23 January 2019 with a second edition on 21 January 2020. It covers four areas: internal governance, the level of human involvement in AI-augmented decisions, operations management, and stakeholder communication.

What is AI Verify?

AI Verify is an AI governance testing framework and software toolkit developed by IMDA. It sets out 11 principles with process checks and evidence, and runs technical tests for fairness, explainability and robustness on traditional machine learning models.

What is the AI Verify Foundation?

It is a not-for-profit body set up by IMDA to harness the open-source community to advance responsible AI testing. It maintains AI Verify and related tools such as Project Moonshot for generative AI.

How do you build an AI governance framework for a company?

List every AI use case, rate each one by probability and severity of harm, decide the level of human involvement, name an owner, and record the PDPA basis for any personal data. Add a staff policy, a vendor file and a review date, and scale the detail to the risk.

Author

Tim Etzal Group

Etzal Group Team, helping businesses grow with technology. Contact us.

Related service

WhatsApp AI Chatbot

We build, train, and run a WhatsApp AI agent that answers customers 24/7, qualifies leads, and follows up.