Singapore Calls for Shared AI Safety Tests and Incident Reporting
Singapore is urging countries to agree on practical safeguards for increasingly capable artificial intelligence, even if their national laws differ. Speaking at the FutureChina Global Forum, Senior Minister of State Tan Kiat How called for shared ways to test advanced models, stronger controls as risks grow and clear channels for reporting serious incidents. He said cooperation on these basics could help keep competition from weakening safety as AI systems gain greater autonomy.
Image: Markus Spiske / Unsplash
Tan said governments could find common ground on how to evaluate advanced models and decide when their capabilities warrant added protection. The tests should consider threats such as cyberattacks, biological misuse and the ability of AI systems to act independently.
Possible controls include restricting access to powerful models, limiting the tools and systems available to them, recording their actions and testing whether safeguards can be bypassed. Human intervention should remain possible when a system behaves unexpectedly.
Singapore has joined more than 20 countries supporting a call for control over frontier AI at the United Nations General Assembly. Tan stressed that cooperation need not mean identical national regulations; it can begin with common technical expectations and reliable ways to exchange information after incidents.
He also argued that Singapore’s advantage lies in putting AI to useful work rather than simply accumulating computing power. As AI agents begin carrying out tasks such as purchases, their identities, permissions and accountability will become more consequential.
Ant International, Mastercard and Visa are working with the Monetary Authority of Singapore on ways to identify and trust AI agents across payment networks. Singapore’s chief AI officer He Ruimin warned that misuse is already a pressing concern and called for several layers of protection, secure digital systems and international cooperation.
Source: ST
