AI may need aviation-style safety regulations to gain trust: Josephine Teo

SINGAPORE - Artificial intelligence may eventually need a safety regime akin to civil aviation, where layers of safeguards and standards give people the confidence to use the technology, said Minister of Digital Development and Information Josephine Teo on Sept 17.

“Trust is an essential requirement for a technology to be regularly relied upon. The safeguards for advanced AI are still taking shape. They will need more research and testing. This takes time,” said Teo in a LinkedIn post.

Her comments on LinkedIn come amid growing debate over whether the rapid pace of AI development should be slowed down to give safeguards time to catch up. Increasingly, developers of frontier AI models are pointing to the risk of losing human control over the technology, which could lead to large-scale cyber attacks and bioterrorism.

Using the analogy of civil aviation, Teo said that a tremendous amount of effort is required to ensure that passengers arrive safely.

Thousands of standards and regulations - from from how the aircraft is manufactured and serviced to how the airspace and airports are managed - keep people safe, she added.

Aviation companies are expected to train their people properly and implement safeguards consistently. Without such assurances, neither passenger nor even crew will board, and civil aviation would never have grown, Teo said.

“In time to come, we will need similar set-ups for AI,” said Teo, adding that the hard and practical work must begin now.

For Singapore, this means building the safeguards alongside adoption, which will entail developing ways to test and assure AI systems.

It also

means strengthening

Singapore’s

technical capabilities to evaluate advanced models, and giving organisations practical guidance on responsible use.

She cited AI Verify, a toolkit developed locally to help companies check if AI results are fair and robust; Singapore AI Safety Institute, a national organization that tests and evaluates artificial intelligence models for safety risks; and the Model AI governance frameworks, which are reference guides to help enterprises deploy AI safely.

Teo said that Singapore will also collaborate internationally because AI safety cannot be tackled by any country alone.

She pointed to the International Scientific Exchange for AI Safety, an annual forum set up by the Infocomm Media Development Authority for international AI safety cooperation. The forum brought diverse perspectives from around the world and helped identify the key priorities that AI safety researchers should look into, set out in a document called Singapore Consensus.

“We will not have all the answers from the onset. But as AI grows more capable, its safeguards must keep pace,” said Teo.

She added that Singapore will continue to test, learn, and strengthen its approach, as trust is core to a vibrant AI hub. The country will continue to work with industry, researchers, and international partners.

The calls for a AI slowdown gathered momentum in recent days after researchers and executives at some of the world’s leading AI companies issued increasingly stark warnings that advances in AI capabilities were outpacing efforts to ensure the technology remained under human control.

Anthropic CEO Dario Amodei said in a lengthy post shared on social media platform X on Sept 12 that AI companies should slow the rate at which they advance model capabilities amid mounting fears of misuse of AI.

In the 3,800-word essay called “We Must Pace The Frontier”, Amodei said that in six to 12 months time, AI agents “could be capable of taking over the entire internet potentially causing hundreds of billions of dollars in damage” and emphasised the need for safeguards.

The call for slowdown was supported by other major AI business leaders like Elon Musk, who runs xAI, and Sam Altman, CEO of OpenAI.

These comments came days after AI researcher Jacob Coxon, who moved from OpenAI to Anthropic, quit the industry, warning that the two US companies were “gambling with our lives” in their race to develop AI models capable of improving themselves.

Such concerns have intensified after recent tests showed that AI models acted beyond the boundaries set by their developers.

In July 2026, OpenAI revealed that its AI models had broken out of their confined testing environment, connected to the internet and hacked Hugging Face, a site developers use to store and share code.

In August, Meta said that one of its models had inadvertently gained internet access and exploited a vulnerability in another company’s systems.

In September, Anthropic disclosed four incidents in which Claude models had gained unauthorised access to third-party computer systems during cyber security tests.

US President Donald Trump rejected calls to slow AI development on Sept 14, arguing that doing so could allow China to overtake the US. He said the US was currently ahead of China in AI and that he wanted to preserve that lead.

China has also pushed back against Amodei’s proposals. Its foreign ministry warned against “fear-mongering” and “vicious competition”, while calling for countries to cooperate on AI governance.