AI Verify Testing Framework and Toolkit

What it is

AI Verify is an AI governance testing framework and software toolkit from Singapore. IMDA and the PDPC launched a minimum viable product on 25 May 2022. The framework and toolkit were open-sourced on 7 June 2023 and are now stewarded through the AI Verify Foundation. An updated framework released on 29 May 2025 covers both traditional and generative AI.

Its purpose is to help an organisation assess how responsibly it implements an AI system, and then show the result to stakeholders. It turns governance principles into things that can be checked.

Who it applies to

It was developed with input from companies including AWS, DBS Bank, Google, Meta, Microsoft and Singapore Airlines. It is open to any organisation that develops or deploys AI and wants a structured way to test and evidence its practices. It is voluntary. In Singapore it complements the Model AI Governance Framework, and the PDPC’s advisory guidelines on personal data in AI systems point organisations to it as a resource. Its mapping to international frameworks makes it usable outside Singapore too.

What it requires

The framework assesses an AI system against 11 internationally recognised AI governance principles.

• Transparency
• Explainability
• Reproducibility
• Safety
• Security
• Robustness
• Fairness
• Data governance
• Accountability
• Human agency and oversight
• Inclusive growth, societal and environmental well-being

It has four elements. Principles state what good looks like. Outcomes describe what each principle should achieve. Processes are the actionable steps, the process checks, an organisation takes to reach those outcomes. Evidence is the documentation that shows each process check was done.

Process checks are published as a PDF, an Excel tracker and a generative AI software tool that produces a summary report. They are complemented by technical testing, through the AI Verify Toolkit for traditional AI and Project Moonshot for generative AI.

The output is a record of what was tested, what was found and what evidence supports it. It does not produce a pass mark or a safety verdict.

Working through the process checks

The process checks are the part most organisations start with. For each process, the organisation carries it out, chooses the type of evidence that shows it (such as an internal policy document), provides that evidence, and marks the process as implemented, not implemented or not applicable. The framework recommends giving reasons for a no or a not applicable answer and noting where the evidence sits for a yes.

The checks are not evenly spread. Safety, transparency, security and accountability carry the most, and explainability and data governance the fewest. Much of the evidence the checks ask for is the ordinary documentation of a governed system, such as policies, model documentation and test results.

Who uses it

The AI Verify Foundation names four audiences. AI application owners and developers use it to document and demonstrate responsible practices. Internal teams use it to check those practices are in place. External auditors use it to validate a client’s implementation independently. Policymakers and standards bodies use it to inform best practice and standards.

Key dates

• 25 May 2022. Minimum viable product launched.
• 7 June 2023. Framework and toolkit open-sourced.
• 29 May 2025. Updated framework released, adding generative AI.

How it relates to other frameworks

AI Verify is the testing counterpart to Singapore’s Model AI Governance Frameworks. Where the frameworks describe practices, AI Verify gives process checks and tests to show those practices are in place. The MAS AI Risk Management Guidelines list the AI Verify Foundation’s initiatives among the national resources financial institutions can draw on.

The AI Verify Foundation publishes crosswalks mapping the framework to the NIST AI RMF and its Generative AI Profile, the Hiroshima Process International Code of Conduct and ISO/IEC 42001. That makes it useful as a way of producing evidence that also speaks to those frameworks, rather than a separate exercise.

A note on this page

This is Aicura’s reading of AI Verify, written to help you understand it. It is not legal advice and it is not the framework itself. Read the primary source from the AI Verify Foundation.