Singapore Announces Updated Model Governance Framework for Generative AI

By‎ Adv. Dotan Hammer

Partner at the Cyber, Privacy & Copyright Group, Pearl Cohen Zedek Latzer Baratz

Reading time: 1 Minute
Save to "My Content"

The Singapore Infocomm Media Development Authority (IMDA) and AI Verify Foundation have announced the publication of the Model AI Governance Framework for Generative AI. Initially published in 2019 and updated in 2020, the current iteration was released for public consultation in January 2024. This version aims to address specific artificial intelligence (AI) risks associated with generative AI, such as hallucinations and copyright infringement. The Framework provides a systematic approach to tackling these issues while encouraging innovation and involving key stakeholders, including policymakers, industry professionals, researchers, and the public.

The Framework outlines nine key aspects –

  • Responsibility is assigned based on the level of control each stakeholder has over generative AI development.
  • The Framework recommends starting with existing personal data protection laws, utilizing Privacy Enhancing Technologies (PETs), and ensuring high data quality.
  • It advises evaluating AI models during training using techniques such as Reinforcement Learning from Human Feedback and Retrieval Augmented Generation, along with benchmarking and red teaming.
  • Establishing processes for timely incident reporting and remediation is suggested.
  • The importance of external audits conducted according to standardized methods to enhance transparency is highlighted.
  • A ‘security-by-design’ approach is advocated, incorporating safeguards such as input filters and digital forensic tools for generative AI.
  • Given the creation of realistic synthetic content, the need for digital watermarking and cryptographic provenance solutions is emphasized.
  • The Framework notes the limitations of current safety techniques and recommends Reinforcement Learning from AI Feedback and post-training model evaluation to ensure alignment and control.
  • Emphasizing the importance of using generative AI to empower the public, it advocates for democratized access to technology, improved public service delivery, workforce development, and sustainability.

Click here to read the IMDA and AI Verify Foundation press release, and here to read the Model AI Governance Framework.