Singapore studying stronger safeguards for high-risk AI: Josephine Teo

Singapore is studying whether new safety measures are needed for high-risk artificial intelligence uses as increasingly autonomous systems create risks that existing safeguards may not fully address. Minister for Digital Development and Information Josephine Teo stated in a written parliamentary response on October 6 that essential services and other higher-risk AI deployments could soon face more rigorous testing, independently verifiable safety evidence, tighter deployment controls, and stronger regulatory oversight.

Singapore Evaluates Multi-Layered Defences Against Autonomous Artificial Intelligence

No single safeguard will be enough to keep pace with potential artificial intelligence threats, according to Josephine Teo, who outlined a multi-layered defence approach in social media posts. This strategy combines stronger cyberdefences, clear limits on what AI can do, appropriate human oversight, and capabilities to test, monitor, and learn as the technology evolves. Singapore’s approach is to deliberately limit what AI agents can do while operating on the presumption that malicious actors will exploit the technology to cause harm more quickly and at scale.

The Cyber Security Agency of Singapore and the Government Technology Agency currently use artificial intelligence to detect vulnerabilities and potential threats more quickly across government systems and critical information infrastructure. Josephine Teo noted that while no unsupervised artificial intelligence agent has launched a cyber attack against government agencies so far, authorities are examining how existing incident reporting channels can be improved to identify artificial intelligence-related incidents and support remediation. Meanwhile, Senior Minister of State for Digital Development and Information Tan Kiat How emphasized during a keynote address at 0G Dev Day 2026 that Singapore provides a connected environment for enterprises to deploy and test emerging technologies before adapting them for other markets across Southeast Asia and beyond.

Frontier Model Access Debated Amid International Safety Commitments

The Singapore government does not currently require frontier artificial intelligence developers to provide access to their models, Josephine Teo confirmed in her parliamentary reply. The Minister explained that a rigid requirement could prove counter-productive if it leads companies to limit their cooperation or information-sharing. Instead, Singapore prefers to build strong collaborative relationships with frontier artificial intelligence labs while drawing on independent research, developer assessments, and international partnerships.

This policy perspective follows Singapore’s recent endorsement of an international call led by Norway and Finland for stronger safeguards around frontier artificial intelligence models. While lawmakers have questioned whether pre-deployment model access should mirror arrangements in the United States and United Kingdom, the ministry maintains that comprehensive risk understanding stems from multiple evaluation sources rather than relying on any single form of access or assessment.

Regulators Test AI Agents in Isolated Sandbox Environments

Authorities have yet to clarify whether any AI company has made its latest frontier model available to a Government agency before deployment, leaving open questions regarding the practical enforcement of recent international safety endorsements. Regulators continue working with industry partners such as Google through the Infocomm Media Development Authority, the Cyber Security Agency of Singapore and the Government Technology Agency to test AI agents in isolated “sandbox” environments using model governance frameworks and starter kits. Whether these voluntary frameworks will suffice as artificial intelligence systems grow increasingly autonomous remains undetermined by the ministry.