Singapore: the toolkit state

Lesson 4 of 5 in Asia-Pacific: Japan, South Korea, Singapore, India, and Australia.

Singapore made a different wager: that the binding-vs-voluntary debate misses the real bottleneck. You can mandate “fairness” and “safety” in statute — but until someone can test for them, the mandate is a wish. So Singapore builds the tests.

The stack began with the Model AI Governance Framework (January 2019, second edition 2020) — Asia’s first detailed, practice-oriented AI governance framework, organised around two anchor principles: AI decisions should be explainable, transparent and fair, and AI systems should be human-centric. When generative AI broke the old assumptions, IMDA shipped the Model AI Governance Framework for Generative AI (May 2024), with nine dimensions: accountability, data, trusted development and deployment, incident reporting, testing and assurance, security, content provenance, safety-and-alignment research, and AI for public good. Both are voluntary. Neither has ever been proposed as legislation — and IMDA says so on purpose.

AI Verify (2022) — the testing framework

A testing framework plus open-source software toolkit that runs technical tests (fairness metrics, robustness checks) and process checks against eleven internationally aligned governance principles, producing a standardised report. It does not certify — it evidences. Handed to the community via the AI Verify Foundation (2023), whose members include the global model providers.

Project Moonshot (2024) — LLM red-teaming

An open-source toolkit for benchmarking and red-teaming large language models: attack modules, scoring, and evaluation runs, aimed at making generative-AI testing reproducible rather than artisanal. Singapore’s answer to the question every regulator now asks: how would we even check?

Global AI Assurance Pilot (2025) — a market for testing

Pairs real deployers with specialist AI assurance firms to test deployed generative-AI applications, seeding a commercial third-party assurance ecosystem — the auditors, methods, and expectations that any future binding regime (anyone’s) would need on day one.

The institutional frame

IMDA (infocomm regulator) drives the frameworks; PDPC issued advisory guidelines on using personal data in AI systems (2024) under the PDPA; the Digital Trust Centre was designated Singapore’s AI Safety Institute (May 2024); National AI Strategy 2.0 (December 2023) sets the ambition. Sectoral regulators bind where it counts: MAS FEAT principles and the Veritas toolkit in finance, MOH guidelines for health AI.

Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.