AI Verify · IMDA Singapore
AI Verify AIVTF v2.0 testing and evidence
11 principles, 62 outcomes, 112 process checks — 104 GenAI-applicable. Official exports, technical tests, and byte-verified attestations on one platform.
AI Verify AIVTF v2.0 is IMDA Singapore's AI Verify Testing Framework for responsible AI, with 11 principles, 62 outcomes, 112 process checks (104 applicable to GenAI), and five technical tests aligned to industry cookbooks.
Key takeaways
- Singapore public sector and IMDA-aligned enterprises use AIVTF for procurement and deployment gates.
- AgenticAssure crosswalks AIVTF to NIST AI RMF and ISO/IEC 42001 — one test run, multiple framework mappings.
- EvidenceExtractionAgent converts model cards and policies into AIVTF console attestations.
AIVTF structure: principles, outcomes, checks
The framework layers governance from high-level principles down to executable checks. GenAI-specific checks cover hallucination, prompt injection, and tool-use risks that classic ML testing missed.
Technical tests and Moonshot cookbooks
Five technical tests map to nine Moonshot cookbooks in AgenticAssure's Benchmark Suite. Red-team results feed directly into AIVTF evidence packages.
Questions compliance teams ask
What is AI Verify AIVTF v2.0?
AIVTF v2.0 is the AI Verify Testing Framework published by the AI Verify Foundation (IMDA Singapore). It includes 11 principles, 62 outcomes, and 112 process checks with 104 GenAI-applicable items.
Who must comply with AI Verify in Singapore?
Government agencies and enterprises aligning to IMDA AI governance guidelines use AIVTF for responsible AI deployment. It is increasingly referenced in public-sector procurement.
Does AgenticAssure export official AIVTF artefacts?
The platform supports official PDF and Annex A CSV exports from test and governance data, plus EvidenceExtractionAgent attestations on the AIVTF console.