Singapore chose not to pass an AI law in 2022. Instead, it built a free toolkit that lets companies test their AI systems and show the results. Four years on, it has become a model other countries study.

The dilemma

By 2022, governments around the world were trying to work out how to govern AI. The European Union was moving toward a law. The United States was issuing guidance. Singapore took a different view. As a small country that depends heavily on cross-border digital trade, it had reasons to be cautious about passing a law too early. Rules written before anyone fully understood the technology risked locking in mistakes, and rigid local rules could have created friction with the many trading partners Singapore depends on.

The regulatory gap

What was needed was a way for companies to show, credibly, that they were using AI responsibly, to their customers, their partners, and regulators in other countries, without yet being forced to comply with a law. Without such a tool, companies were left either disclosing too much by default, or making claims about their AI that no one had any reason to trust.

The approach

In 2022, Singapore's Infocomm Media Development Authority, known as IMDA, launched AI Verify. It is a voluntary testing framework and free software toolkit built around 11 internationally recognised principles for responsible AI. It runs technical tests and process checks on an AI system and produces a report that a company can share with customers, partners, or regulators. In 2023, IMDA made the software open-source and set up the AI Verify Foundation to develop it further. Later that year it added a sandbox for testing generative AI, and by 2025 the testing framework had been extended to cover generative AI tools. The thinking was that a voluntary, open-source, internationally compatible approach would shape company behaviour faster than a law, and would make Singapore a trusted global centre for AI testing.

Results achieved

Take-up grew quickly. By June 2023, IMDA reported that more than 50 companies had shown interest, and the AI Verify Foundation launched with 7 founding members and more than 60 general members. By October 2023, more than 10 major global companies had joined the generative AI testing sandbox. A 2025 Global AI Assurance Pilot involved more than 33 organisations across 9 countries and 10 industries, testing 17 generative AI applications, 16 of which were already in real-world use. The framework is built to line up with the main international AI standards, including the US NIST AI Risk Management Framework, the G7 Hiroshima Code of Conduct, and the ISO/IEC 42001 standard.

Two limits are worth noting. The visible results so far are mostly tools, catalogues, and pilot reports. Whether all this testing actually makes AI safer, makes companies more productive, or protects consumers, in ways that can be measured, is still an open question. The ecosystem is real. The proof that it changes outcomes is not yet in.

The transferable lesson

A voluntary framework only spreads internationally if it works alongside other countries' rules. AI Verify's key design choice was to map its tests directly against the major international standards. That is what turns it from a Singapore initiative into something other countries can use. The lesson for any country building voluntary AI governance is to design for compatibility with international standards from the start, or expect to redo the work later.