Singapore Model AI Governance Framework for Generative AI

What it is

The Model AI Governance Framework for Generative AI is Singapore’s framework for building a trusted ecosystem around generative AI. IMDA and the AI Verify Foundation developed it, and it was released on 30 May 2024 after a consultation on a proposed version issued on 16 January 2024. It follows a June 2023 discussion paper on generative AI and its implications for trust and governance.

Its starting point is that generative AI reinforces some existing AI risks, such as bias, misuse and lack of explainability, and introduces new ones such as hallucination, copyright infringement and value alignment. It argues that no single intervention will be enough and sets out nine dimensions to be looked at together.

Who it applies to

It is guidance, not law, and it is addressed to the whole ecosystem. Model developers, application deployers, policymakers, the research community and the public each have a part. For an organisation deploying generative AI, it is the reference Singapore regulators and customers are likely to use when asking how a generative system is governed.

What it requires

The nine dimensions, in the framework’s own order.

Accountability. Put the right incentives in place so every party in the development chain is responsible to end-users. Generative AI involves several layers in the technology stack, so the framework draws parallels with how responsibility is shared in cloud and software development.

Data. Ensure data quality, for example through trusted sources, and handle potentially contentious training data such as personal data and copyright material pragmatically and fairly.

Trusted development and deployment. Adopt industry best practice in development and evaluation, and give meaningful transparency about baseline safety and hygiene measures, described as “food label” style disclosure. The 2020 framework’s practices for traditional AI are folded into this dimension.

Incident reporting. Establish structures and processes for incident monitoring and reporting, so incidents are notified and remediated in good time and systems improve.

Testing and assurance. Use third-party testing and assurance to show trustworthiness to end-users, and develop common standards for AI testing.

Security. Adapt existing information security frameworks and develop new testing tools for the threat vectors generative models introduce.

Content provenance. Be transparent about where and how content is generated, using technical approaches such as digital watermarking and cryptographic provenance in the right context.

Safety and alignment research and development. Accelerate research into aligning models with human intention and values, with global cooperation among AI safety institutes.

AI for public good. Democratise access, improve public sector adoption, upskill workers and develop AI sustainably.

Key dates

• June 2023. Discussion paper published.
• 16 January 2024. Proposed framework issued for consultation.
• 30 May 2024. Final framework released.

The framework says it will evolve as technology and policy develop. IMDA’s next step was the Model AI Governance Framework for Agentic AI in January 2026.

How it relates to other frameworks

It sits between Singapore’s 2020 Model AI Governance Framework for traditional AI and the 2026 framework for agentic AI. AI Verify supplies process checks and technical testing tools that cover generative AI, and the MAS AI Risk Management Guidelines, which take effect on 7 October 2027, cover generative AI used by financial institutions.

Internationally, the dimensions touch the same ground as the NIST AI RMF’s Generative AI Profile and the EU AI Act’s transparency obligations for AI-generated content. Incident reporting and testing and assurance line up with the operational and performance evaluation clauses of ISO/IEC 42001.

Working with it in Aicura

Singapore framework content is available in Aicura on the Enterprise tier. Once an administrator enables it in Settings under Frameworks, risk and impact assessments for generative AI systems draw on it, and governance policy reviews can be set to apply it. Assessment output is guidance for your judgement. It does not certify a system or declare it aligned with the framework.

A note on this page

This is Aicura’s reading of the framework, written to help you understand it. It is not legal advice and it is not the framework itself. Read the primary source from IMDA and the AI Verify Foundation, and take advice on how it bears on your organisation.