AI Verify Foundation

AI Verify has two parts:

• AI Verify testing framework: an open-source toolkit (on GitHub) that lets enterprises self-assess their AI systems across 11 dimensions (transparency, explainability, repeatability, safety, privacy, robustness, fairness, accountability, human autonomy, well-being, and social benefit)
• AI Verify Foundation: the non-profit organisation responsible for the evolution of governance tooling, ecosystem expansion, and standards-setting

Technically, the AI Verify framework includes:

• Test library: automated test suites (performance, fairness, robustness, etc.)
• Process checklists: standardised questionnaires for human evaluation
• Report generator: outputs standardised compliance reports

The Foundation has 100+ global members, including IBM, Microsoft, Google, Meta, and Salesforce, alongside local players such as Singtel, DBS, UOB, and Standard Chartered.

AI Verify's core innovation is turning AI governance from "principles" into "executable tests".

The pre-existing problem in the field: everyone agreed on "responsible AI", "fairness, transparency, explainability", but no one could tell you how those abstract principles should actually be tested in your system. AI Verify was the first to translate them into concrete pieces:

• 11 evaluation dimensions
• Each dimension mapped to a set of automated tests + manual check items
• Test results generate standardised reports

This approach has been repeatedly borrowed by NIST AI Risk Management Framework, ISO/IEC 42001, and EU AI Act compliance tooling. It's not SOTA technical research, but a "de facto standard" for governance tooling — and victories in standards battles tend to outlast technical victories.

In 2024, the framework was extended to generative AI (Generative AI Verify), adding test modules for LLM-specific risks (hallucination, jailbreak, copyright).

AI Verify Foundation is the smartest move in Singapore's AI strategy.

In the "seven transmission levers" framework:

• Lever 4 (governance): turning the governance framework into a commercially usable tool
• Lever 6 (international): through the Foundation form, internationalising and de-politicising "Singapore governance standards"

A take: Singapore did something with AI Verify that no other country has done — converting a "national governance standard" into a "global open-source tool". The EU AI Act is law: outside EU jurisdiction it has no power. NIST AI RMF is a US official standard: international acceptance is shaped by geopolitics. AI Verify, by contrast, is an Apache-licensed open-source project — anyone can use it without feeling "regulated by Singapore".

This kind of "standards diplomacy" is something only small countries can play — when great powers issue standards, others get defensive; when a small country issues them, they look neutral. AI Verify has secured Singapore the "neutral platform" position in AI governance.

Bottlenecks to watch: the Foundation's funding sustainability (currently relying mainly on IMDA and corporate member fees), ecosystem traction (the member list is glittering but few are deeply engaged), and the pace of technical evolution (how an open-source governance tool keeps up with the rapid evolution of GenAI).