As AI firms call for accountability and limits, an opportunity beckons for Singapore

[SINGAPORE] As artificial intelligence heavyweights overseas call for a slowdown in AI development, citing need for accountability and limits, Singapore has an opportunity to lead in the AI governance space, analysts in the city-state say.

An online essay by Anthropic’s CEO Dario Amodei asking for a slowdown in AI development has stirred debate in the tech community. In his three-step plan, posted on Saturday (Sep 12), Amodei said that the first step is allowing access to third-party evaluators to verify that safety practices and communications are followed, as well as report incidents.

Second is to establish common safety standards as well as limits on the rate of unchecked AI progress. Third is for the US and other democratic governments to coordinate this with authoritarian governments.

“We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain,” said Amodei.

A targeted slowdown in areas where human ability seems to be falling behind in the control of AI is what Professor Bo An, head of the AI division at Nanyang Technological University’s College of Computing and Data Science, supports.

His main concern is how much leeway these AI systems have been given, especially when it concerns access to sensitive data or essential services.

Work on lower-risk useful applications for AI should continue to progress. But if developers cannot contain the system adequately or explain how serious risks are managed, they should pause that activity, said Prof Bo.

“The extra time must deliver better safeguards; a delay alone achieves little,” he added.

Professor Anthony Tung, AI lead, urban sustainability, at the National University of Singapore’s AI Institute (NAII), noted that what is missing currently is rules with teeth. He reckoned that there are no conditions set for when an AI lab should stop, or what evidence will allow it to proceed.

SEE ALSO

‘Whoever wins AI wins’: Trump downplays AI concerns as CEOs call for slowing pace

Anthropic CEO urges slower AI development as Altman, Musk rally behind call

AI-linked Asian stocks slump on calls to slow development; Fed, BOJ hikes loom

AI slowdown calls from Anthropic, OpenAI may weigh on chip stocks, but rally seen intact

More

“My suggestion is that a lab should release a capability only as fast as accredited independent evaluators can assess it, so the pace is set by evaluation throughput rather than by a competitor’s release schedule,” said Prof Tung.

However, Gunasekharan Chellappan, co-chair of industry body SGTech’s AI, Cloud and Data Chapter, is less of an advocate of a slowdown in the pace of development.

His view is that what is needed is to ensure that equal investments are also made in the security and governance models so that they are developing in parallel with AI.

“This cannot be like a nuclear race where, if we win, we would have annihilated ourselves,” he said.

The push instead should be to ensure that all the AI companies implement the necessary safety measures within their own development, Chellappan added.

NAII’s Prof Tung holds similar views, cautioning that obligations should scale with risk rather than be a one-size-fits-all option. If not, regulation might end up entrenching the large frontier labs it is meant to restrain.

“A slowdown observed only by American companies will be fragile. But difficulty in securing universal agreement is not a reason to leave safeguards voluntary,” he added.

Slowing down AI development

Should a slowdown in AI development occur, there will be a need to focus on independent testing, limiting what AI can do and clear human accountability, said Prof Bo. There needs to be explicit conditions set for stopping unsafe experiments, and external evaluators require meaningful access to check for compliance and report incidents.

For organisations deploying AI, this means more controls over money, data and critical systems. There is also a need for tested ways to stop an agent and recover from mistakes, said Prof Bo.

Singapore’s agentic AI framework provides a useful foundation, added Prof Bo. The framework launched in January 2026 provides guidance on technical and non-technical measures needed to deploy agents responsibly.

As part of the framework, organisations are encouraged to assess and cap risks through appropriate use cases and limits on AI agent’s powers and access. Humans should be made accountable for agents by defining checkpoints at which human approval is required. Technical controls and processes throughout an agent’s life cycle should be implemented including testing and controlling access to data and applications.

A unique opportunity

Even as calls for an AI development slow down, businesses are being warned that slowing AI adoption might not be an option for them.

There is still much to gain for businesses. A recent Bain study noted that US$4.7 trillion of profits of global businesses from now to 2035 could be at stake. This is due to the far-reaching impact of AI on the global economy.

The amount dwarfs the estimated US$1.4 trillion in profits that were hit by the last tech disruption, the Internet.

While the Internet reduces the cost of reaching customers, AI reduces the cost of producing the product. This includes analysis, diagnosis, coding and physical work with robots and autonomous vehicles.

Bain estimated that AI will structurally transform 71 per cent of sectors compared with the Internet’s 41 per cent. This includes sectors such as pharma and biotech, healthcare delivery and professional services,

A chance for Singapore to play a role

Amid the calls for slowdown in AI development and need for business to speed up AI adoption, Singapore is in a unique position.

The Republic has been at the forefront of technology adoption, and being a first mover creates an opportunity on the AI front, noted SGTech’s Chellappan. Singapore’s established position in business, financial and legal sectors creates a chance for the country and companies to work on necessary frameworks to be established for AI governance, he added.

Singapore has taken the lead in this regard with AI frameworks such as the AI Verify and AI Tester Accreditation programmes. These frameworks are from the AI Verify Foundation which was launched by the Infocomm and Media Development Authority of Singapore in June 2023.

“But accreditation establishes that a tester is competent within a defined scope; it does not establish that an AI system is safe,” said NAII’s Prof Tung.

There is still an organisation gap where software safety, cybersecurity and AI safety are treated as separate disciplines, funding and regulators, despite incidents increasingly involving all three areas, he added. His view is that Singapore should establish an institute with a mandate to cover all three areas and judge how effective the translation to practice is.

“A slowdown on its own is not a safety policy. Time has to be invested in better evaluation, stronger safeguards and building more resilient systems around AI,” added Prof Tung.