Why Singapore Wants AI Safety to Work More Like Aviation

Singapore’s Minister for Digital Development and Information Josephine Teo sees public trust as a condition for lasting AI adoption. In a September 17 LinkedIn post, she suggested that AI may eventually need safety arrangements comparable to civil aviation’s. Her point was not that AI already has such a regime, but that safeguards must develop as people begin relying on the technology more often. 23

Why aviation is the comparison

Air travellers depend on more than a well-built aircraft. Aviation safety also involves maintenance, airspace and airport management, trained staff, and safeguards applied consistently in day-to-day operations. Teo used those connected responsibilities to illustrate why trust rests on a system of protections rather than any one company’s assurance. 79

The analogy does not mean every aviation rule can be copied into AI. It suggests questions that AI developers, deployers and overseers must answer together: How are systems tested before use? Who checks that they remain safe as they change? What guidance do the people using them need? Teo has described aviation’s approach as multilayered and said Singapore must learn quickly because there is no ready-made AI safety playbook. 5

What Singapore is building now

Teo’s immediate proposal is to build safeguards alongside adoption, while research and testing continue—not to wait for a fully mature rulebook. Reporting on her remarks identifies AI Verify and the Singapore AI Safety Institute among Singapore’s existing safety initiatives. 12

Singapore has also used AI Verify to encourage responsible testing and transparency. Teo has described a programme to accredit independent AI testers so organisations deploying AI can have greater confidence in the people assessing their systems. These are practical steps toward assurance, not proof that AI risks have been settled. 22

Cooperation matters because AI systems and their risks cross borders. Singapore has hosted the International Scientific Exchange on AI Safety, helped develop the Singapore Consensus on Global AI Safety Research Priorities, and participated in joint testing efforts. Those activities support shared research and evaluation; they should not be mistaken for a binding international aviation-style regulator. 181

How this differs from calls to slow AI development

Teo’s emphasis on trustworthy deployment comes amid sharper warnings about increasingly capable AI, including risks involving loss of human control, cyberattacks and bioterrorism. On September 12, Anthropic chief executive Dario Amodei called for pacing improvements to frontier models so safety work and independent verification could keep up. His warning that future AI agents might cause internet-wide disruption was a forecast, not an established outcome. 131011

The policy responses are not identical. Teo’s reported position is to strengthen safeguards while AI use grows; Amodei also wants more time before frontier capabilities advance further. Both approaches turn on the same unresolved question: whether safety practices can keep pace with systems people are being asked to trust. 1311