It is 2028. A Fortune 500 bank in London is deciding where to deploy its first fully autonomous AI agent for credit decisions.
The bank has access to the most powerful AI models in the world. Some come from California. Some come from Beijing. Some come from Paris. The capability is no longer the question. What the bank needs is something else.
It needs an audit framework that satisfies its regulators in three different countries. It needs a testing toolkit that can prove to its board that the model behaves as advertised. It needs a third-party assurance broker whose stamp will be recognized by its corporate counterparties from Frankfurt to Texas.
When the bank’s compliance team finishes its assessment, the recommendation will not name a model provider. It will name a country. And that country, increasingly, is Singapore.
Two AI races
To date, the geopolitical conversation about Singapore’s AI position has been narrated on a single axis. Can Singapore build its own frontier models? Can it match US and Chinese compute scale? Can it retain enough indigenous AI talent to staff national infrastructure as foreign firms move in?
The answers, broadly, are no. Singapore will not produce a new OpenAI, Anthropic or DeepSeek. Its 4,500 current AI practitioners, even tripled to 15,000 by 2029, will not move the global frontier on capability.
But that conversation is being held on the wrong axis. There is a second AI race underway, and on that one, Singapore is already the global leader by a margin that may already be structural.
The first race is about who builds the most powerful AI systems. The second race is about who decides which AI systems are safe to use, in which industries and by whom.
The first race captures the headlines. The second race captures deployment decisions, where most of the enterprise economic value will be created over the next decade.
What Singapore is actually building
On January 22, 2026, at the World Economic Forum in Davos, the Infocomm Media Development Authority launched the world’s first governance framework for autonomous AI agents.
It was the first dedicated rulebook for autonomous AI systems anywhere on earth. It was built in Singapore, by Singapore, for the world to adopt. This is the work the headlines about Mistral and HTX are missing.
Singapore’s AI governance infrastructure now operates at three layers. At the framework layer, the Model AI Governance Framework launched in 2019, followed by a framework for generative AI in 2024, and the world-first framework for autonomous AI agents in 2026.
Each has been mapped to the US AI Risk Management Framework, the European Union AI Act and the international ISO 42001 standard. A company implementing AI Verify, Singapore’s testing toolkit, gets simultaneous credibility with Singaporean, American, and European regulators in a single audit pass.
At the testing layer, AI Verify operates as an open-source assurance toolkit governed by a non-profit foundation owned by IMDA. It convenes a global ecosystem of contributors and is engineered to evolve alongside international standards rather than compete with them.
At the institutional layer, Singapore’s AI Safety Institute leads the ASEAN Working Group on AI Governance. India’s Bureau of Indian Standards has followed Singapore’s lead in adopting ISO 42001 as a national standard. The National AI Strategy 2.0 explicitly names the positioning. Singapore is to become a “trust anchor” in the global AI economy.
The Underwriters Laboratories model
To understand why this position is more valuable than it first appears, consider an unlikely parallel.
In 1894, William Henry Merrill founded Underwriters Laboratories in Chicago. UL did not make electrical products. It tested them. Within decades, the UL mark had become the de facto requirement for any electrical product sold in the United States.
Today, products from over 50 countries are tested and certified by UL each year. A Korean appliance maker cannot ship into US retail channels without it. A Chinese consumer electronics brand cannot list on Amazon without it. UL Solutions reported nearly US$2.9 billion in revenue in 2024.
UL does not make the products. UL certifies them. That position has proven more durable than the position of any single manufacturer. Switzerland built a similar model in finance over the 20th century. Its domestic banks were dwarfed by American, Chinese and Japanese giants.
What Switzerland built was the infrastructure that made trust possible across other people’s capital, with neutrality, regulatory consistency, the power of the Swiss franc and the authority of the Bank for International Settlements (BIS), headquartered in Basel. The result is that Switzerland sits at the center of global financial trust without competing on a financial scale.
Singapore is building the same structural position in AI. Not with the largest models, not with the biggest compute, but with the assurance toolkit that lets foreign AI models be trustably deployed in Singapore, then across ASEAN, then in any market procuring against ISO 42001.
The country has spent 40 years building the credibility this position requires, namely, predictable regulation, low corruption, bilingual access to East and West, a judiciary that international counterparties trust and geopolitical neutrality that, even under pressure, still functions better than most alternatives.
Why the second race may matter more
Now, return to the Fortune 500 bank in London. The bank does not need the most powerful AI. It needs the AI it can defend to its regulators, board and shareholders.
A bank deploying autonomous AI for credit decisions does not need the cleverest model. It needs the auditable one. A hospital does not need the most advanced agent. It needs the certifiable one. A government does not need a homegrown language model. It needs a framework that allows foreign models to be deployed without losing control over their own data or operations.
The trust layer scales globally in a way frontier capability never will. It also captures the bulk of enterprise economic value because it is where deployment decisions are actually made.
The Mistral partnership with HTX, viewed against this background, looks less like dependency and more like a customer relationship. Mistral needs a trusted country to legitimize its enterprise expansion.
Singapore is selling that legitimacy at the framework layer while purchasing model capability at the technical layer, meaning the exchange runs both ways.
The race that is actually happening
The next phase of the AI economy will be shaped by which countries can credibly certify trust at scale.
Building bigger models is becoming easier every year. Building the rulebooks that let big models be trusted across countries is becoming harder, exactly because it requires the kind of slow institutional credibility that cannot be bought or easily replicated.
Singapore has been building that credibility for four decades, well before the AI era began. Its 2026 governance frameworks are the harvest of that long investment.
The world’s regulators are mapping their frameworks to Singapore’s, and the world’s enterprises are increasingly procuring in accordance with the standards Singapore helped harmonize. The world’s first governance framework for autonomous AI agents was written here, not in Brussels or Washington.
Five years from now, when the Fortune 500 bank in London makes its decision, the recommendation will not name California. It will not name Paris. It will not name Beijing. It will name the country whose mark has been on the audit report all along: Singapore.
Chris Chen is an angel investor and founder of Future 500, a Singapore-based founder-led accelerator working with founders to scale beyond their home country.