Josephine Teo’s Case for Aviation-Style AI Safety—and Singapore’s Safeguards So Far

People regularly board planes because they expect a system of standards, trained personnel and safeguards to help them arrive safely. Singapore’s Minister for Digital Development and Information, Josephine Teo, argues that AI will need comparable public confidence as it becomes more capable and widely used. Her September 17, 2026, remarks describe a possible direction for AI safety—not a claim that such a regime already exists. 31924

Why compare AI safety with aviation?

Teo’s comparison is about layers of responsibility. In aviation, trust does not rest on one manufacturer or one safety check. She has drawn a parallel between aircraft makers and AI model makers while stressing that protection extends beyond either of them. For AI, that points toward testing, responsible deployment, trained people and standards that work together, rather than relying on a model’s usefulness alone. 1924

The stakes rise as organisations put AI to work in consequential settings. Teo has raised concerns about loss of human oversight and bias; she has also used AI-assisted credit decisions to illustrate why people need assurance about systems that affect them. Singapore’s AI testing framework has been updated to address risks including sensitive-data leakage, hallucinations and harmful outputs. Separately, reporting on her September remarks identifies concerns about cyberattacks and biological misuse. These are risks to manage, not evidence that each has occurred in a verified incident tied to her proposal. 25262739

Teo has also discussed harmful AI-generated images and obligations for services to remove such content after notification. That is a concrete example of an AI-related harm addressed in her wider safety discussions. The supplied accounts, however, do not establish that a particular recent incident prompted her aviation comparison. 13

What Singapore is building now

Singapore’s approach combines practical assurance with governance guidance. AI Verify provides a framework and toolkit for responsible testing and transparency. Teo has announced a programme to accredit independent, third-party AI testers, intended to give organisations deploying AI more confidence in whom they ask to evaluate their systems. 827

Technical methods are still being developed. A Global AI Assurance Pilot studied how to test the reliability of generative-AI applications; lessons from it informed an IMDA starter kit of testing methods. Singapore has also issued a Model Governance Framework for agentic AI—systems able to act with greater independence—with an emphasis on managing risk and retaining human oversight. 299

Why the work is unfinished

AI products cross borders, so Singapore is seeking evidence-based evaluation methods that can work across different regulatory systems. Teo has said internationally recognised rules will matter but will take time to form. The aviation analogy therefore sets an ambition: build and improve safeguards alongside adoption, while being candid that today’s AI tests and guidance are not yet the equivalent of a settled, comprehensive safety regime. 6730