Minister for Digital Development and Information Josephine Teo told Parliament on Tuesday that Singapore is studying whether high-risk uses of artificial intelligence need stronger safeguards as models become more autonomous, putting banks, telcos and essential services in the frame for tighter oversight.
What Parliament was told
In a written reply to a question filed for the sitting, Mrs Teo said existing frameworks may not fully cover risks from increasingly capable systems that can act on behalf of users. For essential services and other higher-risk deployments, she listed possible measures: more rigorous testing, independently verifiable evidence that safety controls work, tighter limits on when systems go live, and stronger regulatory oversight.
The reply lands as commercial “agentic” tools move from chat windows into workflows that can book appointments, move money and edit records without a human clicking each step. Singapore has promoted AI adoption through the National AI Strategy, but ministers have repeatedly paired that push with sector-specific guardrails in finance and healthcare.
Government systems and reporting channels
Mrs Teo pointed to work by the Cyber Security Agency and GovTech to harden government systems and critical infrastructure. She also said officials are looking at how existing cyber incident reporting channels — through CSA, GovTech and sector leads — can better capture AI-related failures, support remediation and feed back into policy.
On catastrophic or extinction-level risks from frontier models, she said the government does not treat such outcomes as imminent policy drivers, while the National AI Council continues to set broader direction. The council recently backed an international statement signed by about 20 world leaders on managing advanced AI risks.
What changes next for operators
The written answer does not introduce a new Bill or a fixed compliance date. Instead it signals that MDDI and sector regulators may ask operators in sensitive domains to show test results and deployment controls before scaling autonomous features. Banks and payment firms already face MAS technology risk management rules; Mrs Teo’s language suggests AI-specific evidence could be layered on where models make consequential decisions.
For households, the near-term effect is indirect: clearer rules on when an AI assistant may act on a CPF or bank portal without a live human confirmation. Mrs Teo stressed that as AI becomes more capable of acting for people, safeguards must strengthen in step, and the state’s ability to spot emerging risks must improve.
Why the timing matters in Singapore
Local lenders and government agencies have rolled out generative tools for customer service and internal productivity. A parliamentary signal that “high-risk” tiers may need independent safety proof gives compliance teams a reason to slow broad agent rollouts until audit trails and kill switches are documented.
Opposition and back-bench questions on AI have focused less on banning models than on whether Singapore’s light-touch reputation still holds when software can execute multi-step tasks. Mrs Teo’s reply keeps the door open to harder requirements without committing to a single enforcement model — a pattern Singapore has used in data protection and cyber security law.
Businesses waiting for a checklist will need to watch MDDI and CSA circulars. Tuesday’s answer is the policy direction; the operational detail will follow in sector consultations and, where needed, statutory amendments.