NUS Professor on What Anthropic's AI Pause Call Means for Singapore

Anthropic's June report warning that AI systems could soon build smarter versions of themselves has sparked global debate, but NUS Computing professor Jungpil Hahn argues the real question for Singapore is not about pausing development — it is about governing adoption before it quietly reshapes human judgment. In a CNA commentary published Monday, Hahn dissected the gaps in Anthropic's call for a development pause while pivoting to what he calls an overlooked governance blind spot that hits closer to home for ASEAN economies.

Anthropic's "When AI Builds Itself" report flagged the threshold at which AI can autonomously design and build more capable systems without human help. Hahn acknowledges the warning deserves attention, but points out that the proposal lacks specifics on what would trigger a pause, what conditions would lift it, and who would adjudicate. Without those details, he writes, a pause is "merely a deferral that buys no time for actual governance solutions." He also notes that AI training is far easier to conceal than missile silos, and that the strategic incentive to quietly keep building while others pause makes coordinated enforcement nearly impossible under current geopolitical conditions.

The commentary raises uncomfortable questions about the messenger itself. Anthropic confidentially filed for an IPO just days before releasing its pause report, and was simultaneously reported to have embedded engineers with the US National Security Agency for offensive cyber operations using its Mythos model. Hahn argues this contradiction warrants scrutiny: the very actors urging a slowdown have the strongest financial and strategic interests in continuing development, and governance frameworks proposed by incumbent labs must be examined with this in mind.

For Singapore and ASEAN, however, the development-pause debate is not one the region is positioned to drive — neither bloc is a frontier AI developer. But Hahn's central insight is that governing AI adoption matters more than pausing training runs. AI is already embedded in how policymakers research policy, how analysts evaluate evidence, and how the next generation of professionals is trained. Mass AI adoption, he warns, is quietly restructuring the cognitive habits of the people who would staff future governance institutions. "You cannot pause your way to better human judgment," he writes — a line that cuts to the heart of the adoption governance gap.

Why it matters for Singapore: Singapore has positioned itself as a neutral AI hub and a laboratory for responsible governance, from IMDA's new Agentic AI framework to the voluntary AI labelling scheme. But Hahn's analysis suggests these frameworks may not address the deeper risk: that widespread AI adoption could erode the very institutional judgment needed to govern the technology. If policymakers, analysts, and professionals outsource critical thinking to AI systems without maintaining the human capacity to evaluate them, Singapore risks arriving at the global governance conversation having already surrendered the cognitive tools it needs to participate meaningfully. Structuring AI adoption in classrooms, boardrooms, and government offices — not just regulating frontier models — may be Singapore's most urgent AI governance challenge.