Cover Shameek Kundu, executive director of AI Verify Foundation, at the ATxSummit 2025 (Photo: IMDA)
Shameek Kundu, executive director of AI Verify Foundation, at the ATxSummit 2025 (Photo: IMDA)
1 / 1
Drawing from key lessons in its Global AI Assurance Pilot, AI Verify Foundation executive director Shameek Kundu shares insights on laying the groundwork for a trusted AI-powered future
As AI becomes increasingly embedded in our daily lives, trust is essential—not only to ensure safety and accountability, but also to unlock its full potential across sectors, from healthcare and finance to education and public services. Without trust, even the most groundbreaking innovations may face resistance or misuse. Therefore, building a strong foundation is crucial for responsible AI to ensure sustainable and inclusive progress.
This mission is at the core of the AI Verify Foundation (AIVF). “Our primary purpose is looking at how AI can be adopted at scale across society and economy in a reliable way,” says executive director Shameek Kundu. A major barrier to adoption, he shares, particularly for generative AI (GenAI), is the fear among organisations, such as companies, government agencies or nonprofits, that AI systems might fail, hallucinate, generate harmful or embarrassing content, or even cause real harm.
Since its inception in 2023, the not-for-profit subsidiary of the Infocomm Media Development Authority (IMDA) has been working to address these concerns by making it easier for stakeholders to test and evaluate the reliability of their AI systems. As Kundu puts it, “This is part of a much broader discussion around an ecosystem of trust.”
To achieve this, AIVF is building a collaborative community focused on advancing AI testing and governance by developing frameworks, sharing best practices, and driving adoption through advocacy and education. In February this year, in partnership with IMDA, AIVF launched the Global AI Assurance Pilot. This initiative brought together 16 specialist AI testers and 17 deployers of real‑world GenAI applications across 10 industries, including finance, healthcare, human resources, and public and people sectors, in Singapore and beyond.
Read more: Responsible AI pioneer Dr Rumman Chowdhury on what she wants everyone to know about AI
Above The Global AI Assurance Pilot brought together 16 specialist AI testers and 17 deployers of real‑world GenAI applications across 10 industries
The Global AI Assurance Pilot brought together 16 specialist AI testers and 17 deployers of real‑world GenAI applications across 10 industries
1 / 1
A key insight from the pilot was the clear distinction between two types of AI concerns: the high-level, frontier risks often discussed in public discourse—such as deception, existential threats, or misuse—and the practical, day-to-day reliability issues faced by real-world organisations. “While debates around AI safety often focus on models like ChatGPT or Gemini [gaining too much agency], institutions like hospitals, banks or airports are more concerned with whether the systems work reliably, avoid errors, and behave consistently,” says Kundu. These are grounded, operational concerns—whether a model gives the correct output five times in a row, but fails the sixth time—and require a different kind of testing and assurance.
Another major learning was “defining what to test”, Kundu says. Unlike humans, AI systems need clear, codified standards for what constitutes a “good” or “bad” output—and this is not always straightforward. Whether it is capturing medical accuracy or respecting cultural sensitivities, stakeholders must first define the scope of an application, then determine acceptable behaviour within that boundary. Automating tests for these nuanced, often subjective criteria proved to be a complex but critical task.
“The pilot also focused on testing AI in high-stakes applications because these are the use cases where reliability is most critical,” says Kundu. “It’s also important to note that both in Singapore and globally, the use of GenAI in high-risk applications almost always involves human oversight.” For example, the use of GenAI to interpret and summarise colonoscopy and histopathology reports at Changi General Hospital will be conducted under human supervision, ensuring that clinical surveillance recommendations remain firmly under human control. Similarly, even in consumer‑facing tools, such as Changi Airport Group’s chatbot Max, “safeguards are in place to redirect complex queries to human agents”.
In case you missed it: AI or Human? GPT-4.5 proves it’s hard to tell the difference
Above Before joining AIVF, Kundu helped build and scale an AI testing software business at a Silicon Valley start-up (Photo: IMDA)
Before joining AIVF, Kundu helped build and scale an AI testing software business at a Silicon Valley start-up (Photo: IMDA)
1 / 1
These insights shaped the blueprint for the world’s first Testing Starter Kit for GenAI applications, launched this May. This set of voluntary guidelines consolidates emerging best practices for testing large language model-based applications, offering a consistent, structured approach to determining when, what and how to test. It helps organisations identify and assess key risks, such as hallucinations, undesirable content, data disclosure, and adversarial prompts—and provides guidance on how to test for them effectively.
“While AIVF provides a framework for AI testing, we don’t conduct the testing ourselves,” says Kundu. “There’s growing industry demand for formal certification of AI applications, but unlike financial auditing, there’s currently no established accreditation system for AI audits.” In the meantime, AIVF plans to continue running similar programmes, whether a clinic or a sandbox, where companies can present real‑world use cases and be connected with testers. The Testing Starter Kit will also be refined and expanded over time, and the guidelines will be complemented by testing tools to help developers conduct these tests, made available via IMDA and AIVF’s Project Moonshot toolkit.
Read more: AI decoded: what those buzzwords really mean
Above Trust is essential in AI—not only to ensure safety and accountability, but also to unlock its full potential across sectors, from healthcare and finance to education and public services (Photo: IMDA)
Trust is essential in AI—not only to ensure safety and accountability, but also to unlock its full potential across sectors, from healthcare and finance to education and public services (Photo: IMDA)
1 / 1
Kundu is clear on the role of the AI Verify Foundation: “We are an enabler of high-quality testing of AI systems, focusing on three core areas: building community by connecting GenAI developers with testers, supporting the development of guidelines and standards, and providing tools to support testing.” While not a formal standard-setter, the foundation plays a key role in shaping emerging practices. “We also keep a close watch on global developments, engaging with experts and academics to stay ahead in this fast‑moving field.”
Before joining AIVF, Kundu helped build and scale an AI testing software business at Silicon Valley start-up TruEra. He previously served as group chief data officer at Standard Chartered Bank, and has contributed to various AI governance forums, including those of the Bank of England, the Monetary Authority of Singapore, and the OECD and Global Partnership on AI.
Despite rapid technological progress, the insights from the Global AI Assurance Pilot underscore the continued importance of human experts, such as doctors or bankers, at every stage of the AI testing process, from selecting appropriate tests to interpreting results. “Personally, I don’t ever want to live in a world where human judgement doesn’t count,” says Kundu. “The frontier will keep shifting, but I certainly hope there’s always a role for us humans.”
As Senior Editor of Tatler Singapore, Hashirin champions and refines the storytelling across platforms—curating and crafting compelling profiles, cover stories and features that spotlight visionaries shaping culture, business and impact. Driven by curiosity, she draws inspiration from the artists, changemakers and trailblazers she encounters through her work. Beyond the pages of Tatler, she is an avid supporter of local theatre and delights in seeking out art in every city she visits.
