Singapore's call for a UN framework convention on AI safeguards, modeled on the climate-change treaty architecture, lands at a moment when the world's largest AI power is explicitly choosing competition over coordination. The result is a widening vacuum between the pace of AI capability and the pace of rule-making, one that a small state is now trying to fill through the only institution with universal membership.
Why it matters
Singapore's Foreign Affairs Minister Vivian Balakrishnan told the 81st session of the UN General Assembly that "AI capabilities are advancing faster than the human ability to safely manage them"3. The same week, U.S. Rep. Jodey Arrington argued that "erring on the side of innovation and winning the race is the right initial posture" for American AI policy4. ANALYSIS Those two positions, delivered days apart, define the central tension in global AI governance: the country with the most compute power treats regulation as a drag on competitiveness, while a coalition of smaller states treats the absence of regulation as an existential risk.
The big picture
Balakrishnan's proposal is specific. "We must explore all options, including the idea of a UN Framework Convention on AI Safeguards," he said2. A framework convention, as The Business Times noted, could take the form of a broad international treaty establishing foundational principles, cooperative institutions, and architecture for international action, analogous to the UN Framework Convention on Climate Change. Singapore also argued that new international institutions may be needed, ones that "can perform for AI some of the functions that currently exist for setting common technical standards in other domains". The speech called for "rigorous testing and evaluation before deployment" and "clear limits on what autonomous systems can do, along with mechanisms to intervene when they behave in unwanted ways".
Balakrishnan framed the UN as indispensable because it is "the only apex international institution with universal membership". He insisted that "any durable international approach must give all states a meaningful stake, not only those with the most advanced models, the deepest pockets or the largest set of armaments". Over 20 countries and the European Union signed a Finland- and Norway-led declaration on frontier AI model safety on the sidelines of the General Assembly, and Singapore joined that call.
Between the lines
Balakrishnan's speech was notably direct about the US-China dynamic without choosing a side. He said the US has "a longstanding ecosystem edge that quickly converts technology into products, services, companies and profit". He described Chinese AI companies as "driven by cut-throat domestic competition that produces cheap and efficient Chinese AI models" and said China "missed the 20th century industrial revolution and is determined not to miss the AI race again". ANALYSIS By naming both powers' advantages, Singapore positioned itself as a broker rather than a partisan, a posture consistent with its broader multilateralism pitch.
The American counter-argument, at least as articulated at the Texas Tribune Festival, rests on the premise that regulation slows the leader and therefore hands advantage to a rival. Arrington argued the country has "too much at stake with health and safety in the race against China" to risk over-regulation. But Chris Hyams, the former CEO of Indeed, pushed back directly: China has tight regulation around AI use, and "much of that country's innovations are based on U.S. models" because the US has much more compute power. Hyams argued that if the US slows AI development, "China's progress will slow too, like a motor boat pulling a water skier behind it". He also contended that a lack of clear technology regulation stifles companies rather than freeing them5.
ANALYSIS Balakrishnan's rebuttal to the innovation-versus-safety framing was perhaps the sharpest line in the speech: "Rules do not stop or slow progress, but give assurance and confidence to move faster and more effectively for the benefit of all peoples". That argument mirrors Hyams's point from the opposite end of the power spectrum: regulatory clarity is a precondition for durable deployment, not an obstacle to it.
Balakrishnan also raised the starkest version of the safety question: "whether the nuclear button should be in the hands of a human being or an AI," insisting that "human beings must ultimately remain in charge". ◆ By invoking nuclear command authority, Singapore elevated the AI governance conversation from commercial competition to strategic stability, a register that the innovation-first camp in Washington has largely avoided.
What's next
The Finland- and Norway-led declaration, now backed by over 20 countries and the EU, provides a near-term coalition for advancing frontier-model safety norms. Singapore said it would "continue to contribute through platforms such as the digital pillar for the Forum of Small States" and by making its governance frameworks and testing tools available to all. Whether the framework-convention concept gains traction depends on whether major AI powers engage or treat it as a forum they can afford to ignore. Balakrishnan's own assessment of the stakes was blunt: "The work of building trust and laying the ground for global cooperation must begin now"1.