Singapore AI Verify — the IMDA testing toolkit alignment.
How Hael produces the AI Verify governance attestations and testing-toolkit alignment records MAS-regulated firms are increasingly required to surface.
What AI Verify provides
AI Verify is Singapore's voluntary AI governance testing framework, launched by the Infocomm Media Development Authority (IMDA) in 2022 and extended to generative AI in 2024. It comprises the Model AI Governance Framework — eleven principles covering internal governance, decision-making, operations management and stakeholder engagement — and the AI Verify Testing Toolkit, which provides standardised technical and process tests for AI system trustworthiness.
AI Verify is voluntary but increasingly procurement-relevant. Singapore Monetary Authority (MAS) regulated financial institutions are referencing AI Verify alignment in supervisory engagement; Singapore government procurement increasingly scopes AI Verify testing as a vendor expectation; ASEAN cross-border data flow frameworks reference AI Verify as a governance benchmark. The Project Moonshot extension addresses generative AI specifically with red-teaming, prompt-based evaluation and content safety testing.
How Hael runs it
Hael produces the AI Verify alignment artefact — mapping each Model AI Governance Framework principle to operational controls running on the platform. Internal governance pulls from policy library and governance decision records. Decision-making pulls from oversight assignment and FRIA. Operations management pulls from lifecycle state machine, evaluation pipeline and incident response. Stakeholder engagement pulls from consultation records.
AI Verify Testing Toolkit alignment is operationalised through the platform's evaluation pipeline. Technical tests (fairness, robustness, explainability) and process tests (governance, accountability, human oversight) are wired as evidence collectors. Test outputs are sealed in the audit chain. The Project Moonshot extension for GenAI — red-teaming, prompt-based evaluation, content safety — is supported through the platform's evaluation infrastructure for generative agents.
Questions
Is AI Verify mandatory for MAS-regulated firms?
AI Verify is voluntary, but MAS supervisory engagement increasingly references AI Verify alignment for AI deployments in regulated financial activities. MAS Notice on AI/ML model risk management interacts with AI Verify principles. Practical effect: AI Verify alignment is becoming an expected component of supervisory readiness.
How does AI Verify relate to the EU AI Act and NIST AI RMF?
AI Verify shares the trustworthiness frame with both. Mapping is straightforward: AI Verify Model AI Governance Framework aligns with NIST AI RMF GOVERN/MAP functions; the Testing Toolkit aligns with NIST MEASURE; Project Moonshot aligns with EU AI Act GPAI provider obligations under Article 53/55.
What does Project Moonshot add for generative AI?
Project Moonshot extends AI Verify with red-teaming methodology, prompt-based evaluation, content safety testing and generative-specific governance principles. It is the substantive technical surface for testing GenAI systems against AI Verify trustworthiness criteria.
See Hael produce your AI Verify alignment.
Model AI Governance Framework mapping, Testing Toolkit alignment, Project Moonshot evaluation — sealed with hash-chained provenance.