Detection Without Legitimacy: Three Models of AI Governance in India, Singapore, and Malaysia
A forensic AI tool can tell you a video is fake. It cannot tell an institution what to do about that finding. This is the gap that kept surfacing in conversations I had this summer with researchers across Malaysia and Singapore — at ISIS Malaysia's Centre for Responsible Technology, the Khazanah Research Institute, and the National University of Singapore. What surprised me was not disagreement over how dangerous synthetic media is. It was near-total agreement that technical detection, however accurate, cannot by itself decide who gets to declare something false. That decision belongs to institutions — courts, regulators, platforms — and India, Singapore, and Malaysia are each answering the question of institutional legitimacy differently, in real time, right now.
This is the real comparison worth making. Not which country has the most advanced AI policy, but which model of legitimacy each is building — and what each model's blind spot reveals about the others.
Singapore: Legitimacy Through Proximity
Singapore has become one of the region's most closely watched AI governance laboratories, through instruments like the Model AI Governance Framework, the AI Verify testing toolkit, and, in January 2026, the world's first governance framework for agentic AI. But what makes these instruments work is not technological sophistication — it's institutional proximity. Singapore is compact enough that the regulator, the researcher, and the company being regulated often know each other personally. Legitimacy here is manufactured through relationships as much as through statute.
This proximity gives voluntary frameworks real teeth: companies face unusually high reputational visibility in a policy ecosystem this small, so quiet non-compliance is costly even without legal force. But the same feature is Singapore's structural blind spot. Legitimacy-by-proximity depends on a density of trust that cannot be exported or scaled. It works when the actor being governed is a known, cooperative institution. It strains against adversarial, anonymous harms — deepfakes, coordinated disinformation campaigns — where the bad actor was never in the room of mutual acquaintance to begin with. Singapore's own state-backed labs are now building deepfake detection and provenance tools precisely because proximity alone cannot catch what it never had visibility into.
Malaysia: The Case for a Reluctant Middle
If ASEAN produces a country capable of anchoring a bloc-wide AI safety standard, the strongest candidate is not Singapore. It's Malaysia — precisely because its position is legible to the rest of the bloc in a way Singapore's isn't. Singapore's institutional capacity is real but so far removed from most of ASEAN that other member states struggle to see their own constraints reflected in it. Malaysia occupies a harder, more transferable middle: serious research infrastructure found in KRI or ISIS Malaysia alongside the same coordination gaps most of the region shares — jurisdiction split across communications regulators and data protection bodies, and a public conversation on AI harms still taking shape.
That middle position is becoming concrete. Malaysia's National AI Office held a public consultation on a proposed AI Governance Bill in July 2026, and by September the Digital Ministry said the draft was ready to go to Cabinet. It sits alongside a National AI Action Plan built to interoperate with ASEAN's Digital Economy Framework Agreement — legislated through a central coordinating office working across existing sectoral regulators, a structure far more replicable elsewhere in the bloc than Singapore's density-dependent model.
One data point is worth sitting with. Khazanah Research Institute's governance research found that when Malaysian stakeholders ranked AI-related risks, data governance and cybersecurity dominated the responses, while misinformation and trust in AI ranked surprisingly low. This isn't evidence that misinformation is a minor risk — it's evidence of a governance-attention mismatch. A researcher I spoke with in Kuala Lumpur framed it plainly: the institutional muscle memory in the region was built for the risks that arrived first, not the risks arriving now. A skeptic might argue Malaysia's middle position also means it lacks the decisive institutional capacity to move fast — and that's fair. But slower, more legible legislation that other ASEAN states can actually copy may matter more than just speed alone.
India: Regulation by Accumulation, Hardening Fast
India has not built a single dedicated AI law — it governs through accumulation: the Digital Personal Data Protection Act, MeitY advisories, and, as of November 2025, the India AI Governance Guidelines, which created an inter-ministerial AI Governance Group, a Technology & Policy Expert Committee, and an AI Safety Institute. For a country at India's scale, this is not obviously a weakness — waiting for a comprehensive AI statute would mean regulating nothing while harms compound in the meantime.
What distinguishes India is how fast one piece of that accumulation has hardened into binding law. On 10 February 2026, MeitY notified amendments to the IT Intermediary Guidelines specifically addressing "synthetically generated information." Platforms must label confirmed synthetic content and embed permanent, non-removable metadata identifying the platform used to create or alter it — a mechanism for retrospective attribution, not a demand to break end-to-end encryption, a distinction worth being precise about given India's history of encryption disputes. Non-compliance risks losing Section 79 safe-harbour protection entirely. This is a materially different instrument than a voluntary framework, and it arrived faster and narrower than a general AI Act would have.
The accumulation model's characteristic failure remains regulation by crisis rather than design. But it can also produce fast, targeted hardening exactly where harm is most acute — arguably a more honest response to synthetic media right now than Singapore's continued reliance on voluntary compliance or Malaysia's still-drafting statute. The open question India hasn't answered yet: metadata and labels establish that content is synthetic, but neither its 2026 amendment nor the broader evidentiary framework yet settles how a court should weigh a probabilistic forensic verdict — from a tool like a MesoNet or XceptionNet classifier — as proof. Detection is being legislated faster than the legal weight of detection itself.
Why the Region Can't Simply Harmonize
All three gaps point to the same underlying problem: detection capability is outpacing institutional capacity to make detection matter. And this is also why ASEAN, as a bloc, has struggled to move past guidelines into binding regional law. ASEAN is not the EU — it has no supranational body that can impose a directive the way Brussels can. Its founding logic runs on consensus and non-interference; member states retain enforcement sovereignty in a way EU members formally cede in specific domains. The ASEAN Guide on AI Governance and Ethics may not be a first step toward eventual hard law so much as the practical ceiling of what the bloc can produce collectively. If binding regulation comes to this region, it will likely come from individual states first — which is exactly why comparing India, Singapore, and Malaysia as separate live experiments matters more than waiting for one ASEAN-wide answer.
Conclusion
None of these three models is a template for the others to copy wholesale. Singapore's proximity-based trust cannot be manufactured elsewhere. Malaysia's middle position is only an asset if it chooses to lead rather than wait to be told what a mature framework looks like. India's scale is only an advantage if accumulation eventually resolves into coherence rather than permanent patchwork. What ties the three together is a harder claim than "regulate AI well." It's that the central problem of AI governance is no longer whether a machine can detect what's synthetic. It's who has the institutional standing to decide what counts as real — and until that question is settled, better detection tools will keep arriving faster than the legitimacy to act on what they find.
Sources
- Personal Data Protection Commission Singapore, Model AI Governance Framework.
- AI Verify Foundation, AI Verify.
- IMDA, Model AI Governance Framework for Agentic AI, launched 22 January 2026.
- National AI Office Malaysia, consultation paper on the proposed AI Governance Bill, 10–31 July 2026.
- Ministry of Electronics and IT, IT (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, notified 10 February 2026.
- Ministry of Electronics and IT, India AI Governance Guidelines, November 2025.
- ASEAN, ASEAN Guide on AI Governance and Ethics, 2024.
- Khazanah Research Institute, governance research on stakeholder rankings of AI risks.
Ruwala, Mihir. “Detection Without Legitimacy: Three Models of AI Governance in India, Singapore, and Malaysia.” SSRN, posted 20 November 2026. doi.org/10.2139/ssrn.7707818. Independent, not peer-reviewed. CC BY 4.0.