Overview of Singapore’s Model AI Governance Framework for Generative AI
Singapore’s Model AI Governance Framework for Generative AI (MGF for GenAI) is a voluntary yet influential framework designed to foster a trusted environment for the development and deployment of generative AI. Unlike traditional AI governance models, this framework specifically addresses the unique risks and opportunities presented by generative AI systems, such as large language models (LLMs). The framework is maintained collaboratively by the Infocomm Media Development Authority (IMDA) and the AI Verify Foundation, reflecting a commitment to a balanced and systematic approach to AI governance.
The framework does not impose legally binding obligations, as Singapore lacks a horizontal AI statute. Instead, its guidance aligns with existing regulatory requirements under data protection, online safety, and sector-specific laws, ensuring compliance without duplicative burdens. Its voluntary nature encourages industry adoption while fostering innovation in a responsible manner.
Nine Dimensions of the MGF for GenAI
The MGF for GenAI outlines nine key dimensions to create a comprehensive governance structure for generative AI. These dimensions are designed to address risks holistically while enabling innovation. The framework builds upon the earlier Model AI Governance Framework for Traditional AI, released in 2019 and updated in 2020, by incorporating new considerations specific to generative AI.
- Incentive Structures for Responsibility: Establishing mechanisms to ensure accountability across the AI development lifecycle, from model creators to deployers, with a focus on end-user trust and safety.
- Data Quality and Training Data Management: Addressing challenges related to the sourcing, quality, and ethical implications of training data, which is foundational to generative AI models.
- Baseline Safety and Hygiene Measures: Promoting transparency in development, evaluation, and disclosure processes, aligned with industry best practices to mitigate risks before deployment.
- Emerging Threat Vectors: Proactively identifying and mitigating new risks introduced by generative AI, such as deepfakes, misinformation, or unintended biases.
- Content Provenance Transparency: Enhancing clarity about the origins of AI-generated content to provide useful signals for end-users, improving trust and accountability.
Development and Evolution of the Framework
The proposed framework sought to provide practical guidance for model developers and policymakers, emphasizing a collaborative and inclusive approach. Feedback from stakeholders, including industry experts and researchers, played a crucial role in refining the framework before its final release on May 30, 2024.
The earlier Model AI Governance Framework for Traditional AI, released in 2019 by IMDA and the Personal Data Protection Commission (PDPC), laid the groundwork for this evolution. The second edition of the traditional framework, published in 2020, introduced additional considerations such as robustness and reproducibility, further refining its applicability. The MGF for GenAI continues this trajectory by addressing the distinct challenges posed by generative AI, including scalability, interpretability, and ethical deployment.
International Collaboration and Future Directions
The framework is designed to facilitate global conversations among policymakers, industry leaders, and researchers, aiming to establish a trusted development environment for AI on an international scale. By promoting collaboration, the MGF for GenAI seeks to harmonize governance practices and foster innovation while mitigating risks.
Future efforts under the framework will focus on developing detailed guidelines and resources for each of the nine dimensions. These resources will provide actionable steps for organizations to implement the framework’s principles, ensuring a systematic and balanced approach to generative AI governance.
Required actions
- Nine dimensions: accountability, data, trusted development and deployment, incident reporting, testing and assurance, security, content provenance, safety and alignment research, and AI for the public good - Organisations are encouraged to test with AI Verify and adopt provenance and incident-reporting practices