Singapore governs AI through voluntary guidance, not an AI law. IMDA's Model AI Governance Framework covers traditional, generative and (since Jan 2026) agentic AI. AI Verify provides a testing framework and toolkit. Here is how it maps to ISO 42001, NIST AI RMF and SOC 2.
LowerPlane serves startups in the United States, Singapore, India and worldwide.
Original Model AI Governance Framework. Nine dimensions covering governance, operations, stakeholder interaction, and ongoing review.
Guidance for generative AI deployers across 9 updated dimensions — safety, provenance, incident reporting.
Launched 22 Jan 2026 by IMDA, updated with case studies in May 2026. Addresses autonomous, multi-step AI agents in production.
On 14 Oct 2026, the AI Verify Foundation and IMDA released an Agentic Testing Starter Kit and accredited the first independent testers for agentic systems. For AI startups shipping autonomous or multi-step agent features, Singapore now expects third-party testing evidence.
| Theme | MGF / AI Verify | ISO 42001 | NIST AI RMF | SOC 2 |
|---|---|---|---|---|
| Governance & policy | MGF Governance | ISO 42001 Cl. 5–6 | NIST AI RMF Govern | SOC 2 CC1/CC2 |
| Risk management | MGF Risk Management | ISO 42001 Cl. 8 | NIST AI RMF Map/Measure | SOC 2 CC3 |
| Explainability / robustness | AI Verify tests | ISO 42001 Annex B | NIST AI RMF Measure/Manage | — |
| Monitoring / incident response | MGF Operations | ISO 42001 Cl. 9 | NIST AI RMF Manage | SOC 2 CC7 |
Separately, CSA Cyber Essentials and Cyber Trust (2025) include an optional AI security pillar — focused on the security of AI systems in operation.
No. Singapore governs AI through voluntary guidance rather than an AI law. IMDA's Model AI Governance Framework and AI Verify are reference frameworks that enterprise buyers and regulated-sector customers increasingly ask vendors to adopt.
AI Verify is a testing framework and toolkit from the AI Verify Foundation (Singapore). It provides a process-based framework plus a software toolkit for third-party AI governance testing, used by developers and deployers to evidence responsible AI practices.
Yes. AI Verify Foundation has published a crosswalk between AI Verify and the NIST AI Risk Management Framework. Teams running NIST AI RMF or ISO 42001 can use the crosswalk to align their evidence with Singapore's framework expectations.
You are still a deployer under MGF/MGF-GenAI, and buyers evaluate your governance of the system as deployed. You get to lean on the LLM provider for base-model controls (reflect their DPA and sub-model attestations in your evidence), but prompt-logging, model-risk assessment, change management and user-safeguards remain your responsibility.
Government procurement, regulated-sector buyers (healthcare, finance) and large enterprises undertaking AI-specific vendor reviews are the most common askers. AI Verify toolkit output is becoming a standard attachment to vendor questionnaires alongside the SOC 2 report.
No. Membership is optional and signals public endorsement of responsible AI. The toolkit is openly available, and non-members can run tests and submit results. Membership can be useful for vendor directory visibility.
The Oct 2026 Starter Kit provides scoping templates, test categories and reporting templates specific to agentic AI (autonomous, multi-step systems). It is intended for use with the five accredited independent testers. For AI startups shipping agent features, expect enterprise buyers to request Starter Kit output or an accredited-tester report in 2026-2027 reviews.
LowerPlane runs ISO 42001 alongside your SOC 2 or ISO 27001 program with the Singapore crosswalk pre-built.