AI industry can learn from aviation and nuclear safety: Experts

SINGAPORE – If artificial intelligence gets aviation-style regulations, this could mean having an international governing body and national regulators ensuring compliance with a set of globally agreed rules, said industry players.

Jonathan Zhang, president of the Singapore AI Association, said: “The opportunity for AI is that it can compress a journey that took aviation most of the 20th century into a much shorter timeframe because we already know what the destination looks like – internationally trusted, independently verifiable safety assurance.”

While there are concerns that regulations could slow the progress of AI innovation, safety should not be seen as an impediment, said other experts, adding that the AI industry could also learn from regimes that manage nuclear weapons and financial risks.

Hassan Shahidi, president and chief executive of US-based non-profit Flight Safety Foundation, said: “A strong safety framework can enable innovation because society is more willing to adopt new technology when it has confidence that the risks are being responsibly managed.”

On Sept 17, Minister for Digital Development and Information Josephine Teo said that AI may eventually need a safety regime similar to that for civil aviation to foster trust.

Although more research and testing are still needed, Singapore must start building safeguards alongside the adoption of AI now, she added.

Her LinkedIn comments came after recent calls – and opposition – to slow the rapid pace of AI development, amid debate over potentially catastrophic risks that the tech could pose.

One of aviation’s most important achievements was the establishment of internationally accepted rules for accident investigation.

The Paris Convention of 1919 established some of the first international aviation rules.

Later, the 1944 Chicago Convention created the framework that led to the formation of aviation’s global governing body in 1947, the United Nations’ International Civil Aviation Organization (ICAO).

Aircraft were operating in international airspace before today’s international rules existed.

During the interim period, aircraft were governed by national laws, bilateral agreements and earlier international conventions.

“Under today’s ICAO framework, the purpose of a safety investigation is to determine what happened and prevent recurrence, rather than to assign blame or liability,” said Shahidi.

“That learning culture is particularly relevant to AI.”

Another aviation principle that translates well for AI is rigorous testing before deploying high-risk systems, he added.

“Do not wait for a catastrophe to build the safety architecture,” Shahidi urged, explaining that aviation learnt through accidents.

Singapore has a head start in testing, having launched in 2022 the AI Verify toolkit that runs technical audits to check whether AI results are fair, robust and can be explained.

It also established the Singapore AI Safety Institute, formally designated in 2024, to test and evaluate AI models for safety risks.

Dennis Yap, founder and CEO of home-grown start-up AI Seer, said having this testing infrastructure is important to prepare for the enforcement of global regulations, standards and agreements.

“Singapore also has credibility with both the US and China, so we can be positioned as a neutral and trusted party where the world goes to get AI checked,” Yap added.

Safety first?

Chew Hwee Yong, the CEO of the Association of Aerospace Industries (Singapore), said safety is embedded in the design, manufacture, maintenance and operation of an aircraft or its components.

Alan Tan, a professor who specialises in aviation law and politics at the National University of Singapore, said: “No one disputes the need for rigorous aircraft safety.”

But for AI, there is little agreement on the extent to which safety should be prioritised, he added.

While leaders of major tech companies – Anthropic CEO Dario Amodei, OpenAI boss Sam Altman and SpaceXAI owner Elon Musk – have called for a slowdown in AI development, US President Donald Trump has rejected these calls, arguing that doing so could allow China to overtake the US.

China also pushed back and warned against “fearmongering” and “vicious competition”, following comments by Amodei to “pace the frontier” and take measures to slow China’s progress in AI.

Tan said the lack of a consensus means that the US and China – as the main purveyors of AI tech – need to take the lead on developing the necessary AI safeguards.

Trump will meet Chinese President Xi Jinping on Sept 24, and AI is expected to be on the agenda.

Fierce AI competition between the two countries could, however, limit the outcomes.

Even so, in the lead-up to the meeting, the US has said it is open to discussing shared risks with China during the AI talks this weekend.

“In time, a UN-sponsored treaty and agency could be created – akin to ICAO in aviation – to bring in other countries to pursue AI safety mechanisms,” said Tan, adding that setting up an ICAO-like body for international discussions on AI will be a priority.

Recognising that AI safety cannot be tackled by any country alone, Singapore has been engaging with other nations through an annual forum, the International Scientific Exchange on AI Safety, since 2025.

Efforts like this and Singapore’s AI testing infrastructure are essentially doing in a few years what took aviation decades, noted the Singapore AI Association’s Zhang.

“That’s why we’re optimistic real alignment on global AI rules is achievable within this decade, rather than across generations,” he said.

Besides aviation, observers point to other models for how AI safety rules could develop, such as nuclear non-proliferation pacts.

This includes the way the US and the former Soviet Union engaged bilaterally through the Strategic Arms Limitation Talks during the Cold War to limit and reduce strategic nuclear weapons, said NUS’ Tan.

A multilateral Treaty on the Non-Proliferation of Nuclear Weapons was also introduced to bring in other states.

Tan said it is essential for the superpowers to start talking and working on something similar for AI.

“Without their involvement and commitment, any aim to establish AI safeguards and a global consensus is bound to fail,” he added.

AI Seer’s Yap said that the AI sector can learn from the financial sector.

The Monetary Authority of Singapore, for instance, does not regulate neighbourhood remittance shops the way it regulates banks.

He suggested that large AI players carry the heaviest obligations.

These could include conducting independent tests before a product is released, reporting any data leaks or hacking incidents, demonstrating that AI will not harm the public, and footing most of the testing bill.

Smaller AI companies may face fewer obligations so they are not snuffed out.

“Aviation itself is tiered this way,” Yap added.

“The rules for a small radio-controlled plane should not be the rules for a large A380 jet.”