Singapore pushes for global AI safeguards as governance struggles to keep pace

Singapore has called for a stronger international framework to govern the risks of artificial intelligence, arguing that the rapid development of AI has outpaced the global systems needed to manage its risks.

Speaking at the 81st session of the United Nations General Assembly, Singapore Foreign Affairs Minister Vivian Balakrishnan proposed exploring a UN Framework Convention on AI Safeguards and potentially creating a new international institution focused on AI standards and verification. 

The proposal comes as governments face a growing challenge: how to allow AI development to continue while establishing safeguards that work across borders.

For businesses and HR leaders, the discussion is increasingly relevant as AI moves from experimentation into workforce deployment, decision-making and enterprise operations.

AI governance cannot stop at national borders

One of the central messages in Balakrishnan's speech was that AI risks are inherently cross-border. He identified three broad categories: the loss of human control over autonomous systems; the potential misuse of AI by rogue actors to develop biological weapons or other weapons of mass destruction or disruption; and wider economic, social and political disruption.  

That makes fragmented national regulation potentially insufficient, particularly as advanced AI models can be developed in one country, deployed globally and used by organisations and individuals across jurisdictions.

Singapore's proposal therefore centres on creating a shared baseline for how AI systems should be tested, evaluated and controlled.

Balakrishnan argued that AI development should not be halted. Instead, safeguards should develop alongside capabilities.

He compared AI to a high-performance car: a more powerful engine requires better brakes, seat belts and common traffic rules if it is to be used safely.

Testing could become a central pillar of AI governance

A key takeaway from Singapore's position is the emphasis on testing and evaluation before deployment. Balakrishnan called for rigorous assessment of AI systems, clear boundaries around what autonomous systems can do and mechanisms that allow humans to intervene when systems behave unexpectedly. 

This could have implications beyond governments and frontier AI developers.

As organisations increasingly deploy AI in recruitment, performance management, learning, workforce planning and other people processes, comparable testing and evaluation standards could eventually become an important part of enterprise AI governance.

Singapore is also advocating for greater cooperation on how AI capabilities and risks are measured. Balakrishnan suggested that countries could work towards comparable testing methodologies and mechanisms for rapidly reporting serious AI incidents across borders.

Such mechanisms could help move AI governance away from broad principles towards more operational safeguards.

The bigger challenge may be trust

Singapore's proposal also highlights a problem that is harder to solve through regulation alone: lack of trust between countries.

AI has increasingly become part of strategic competition, particularly between major technology powers. Balakrishnan argued that any international framework would need to give countries a meaningful stake, rather than being shaped only by nations with the most advanced models, financial resources or military capabilities.

For a global framework to work, countries would therefore need to agree not only on what constitutes an unsafe AI system, but also on how information about risks, incidents and testing results is shared.

Singapore's proposed starting point is scientific cooperation. Even when governments disagree over regulation, they could potentially collaborate on understanding AI capabilities, developing common testing approaches and establishing mechanisms to report serious incidents.

Singapore is not starting from zero

The proposal would build on several existing international initiatives rather than creating an entirely new governance architecture from scratch.

Balakrishnan pointed to the UN's Independent International Scientific Panel on AI, the Global Dialogue on AI Governance and the International Telecommunication Union's AI for Good Global Commission as existing platforms that could contribute to greater international coordination. 

Singapore has also joined a broader call for greater safety and security of frontier AI models, alongside more than 20 countries and the European Union. The declaration calls for international government oversight of frontier AI rather than relying solely on corporate self-regulation. 

The proposed UN convention would therefore represent a possible next step in a governance conversation that is already underway.

Why a new international institution is being considered

Singapore's proposal goes beyond common principles and raises the possibility of a dedicated international body for AI.

Balakrishnan pointed to the roles played by existing institutions such as the International Telecommunication Union, which helps establish technical standards, and the International Atomic Energy Agency, which provides verification functions in the nuclear field. 

The idea reflects a broader question facing AI governance: whether existing institutions can absorb the rapidly evolving technology or whether AI eventually requires its own permanent international architecture.

Singapore's position is not that AI development should be paused. Instead, it is that the systems surrounding AI need to become more sophisticated as the technology becomes more capable.

That means organisations may increasingly need to consider questions such as:

• How has an AI system been tested before deployment?

• What decisions should remain under human control?

• What happens when an AI system produces an unexpected outcome?

• How are serious incidents identified and reported?

• Can AI risks be assessed consistently across markets and jurisdictions?

These questions are particularly relevant as AI moves deeper into workforce and business processes.

The direction of travel suggested by Singapore is therefore less about choosing between AI innovation and AI regulation, and more about building the governance infrastructure that allows both to develop together.

As Balakrishnan put it, common rules and safeguards should not necessarily prevent progress; they can provide the assurance needed to use increasingly powerful technologies at scale.