Who should own AI? Lawsuits against and for nationalization

In May, DBS chief economist Taimur Baig and former International Monetary Fund economist and author Anthony Annett argued that: business times Nationalizing parts of the AI ​​industry would be treated similarly to nuclear weapons and central banks.

It is a well-argued piece and requires a careful response.

The question this issue raises is not really “should we nationalize AI?” Because that framework is too obvious. AI is not one thing. This is a foundational model trained in thousands of texts, chip fabs, data centers, software companies creating AI agents, and startups building applications around the world.

The real question is what, exactly, and how much to nationalize.

nationalization lawsuit

Baig and Annette’s discussion is worth hearing in full, as it addresses some legitimate concerns.

Their primary concern is safety. In the wrong hands, advanced AI systems could disrupt financial systems, erode or destroy wealth, and interfere with public utilities such as power grids, water systems, and transportation networks. Fraudsters could use AI to spread disinformation on an industrial scale, run frauds that are indistinguishable from genuine communications, or worse, engineer deadly pathogens.

These are not science fiction scenarios. These are concerns expressed by the very institutes behind the technology.

Baig and Annette reach for the nuclear analogy to further their point. In addition to providing a clean and reliable energy source, it can also produce weapons of mass destruction. This dual-use nature is why nuclear technology is everywhere under state control and bound by international non-proliferation treaties.

The argument is that if AI poses comparable dual-use risks, then perhaps it deserves equal treatment.

Their second concern is power. A handful of companies, including OpenAI, Anthropic, Google, and several others, are building the foundational systems on which entire economies may soon run. In addition to exacerbating inequality, such concentration is also economically unsustainable. It could be translated into political power, and public ownership would blunt that translation.

There is also a discussion about fairness. The breakthroughs that underpin today’s AI date back to decades of publicly funded research, and the models themselves are trained on data effectively generated by all of us.

If the public financed the seeds and provided the soil, perhaps it deserves a share of the harvest, a capital dividend, to be distributed in the way sovereign wealth funds distribute profits, rather than a windfall captured entirely by a handful of shareholders.

It’s a convincing case. However, it is too broad to know how AI actually works.

The nuclear analogy doesn’t really apply. Uranium is a physical substance. It can fence reactors, guard enrichment facilities, and inspect nuclear facilities to detect diversion to weapons-grade uranium. This is essentially the role of the non-proliferation regime.

AI isn’t like that. It’s code, it’s learning models, it’s know-how. It has already spread to thousands of researchers, universities, open source repositories, and is now spreading across borders that no treaty currently governs, as seen with the rise of the freely downloadable Chinese model. You can’t ring-fence something that’s already leaked to millions of laptops.

Even if the focus is narrowed down to a small number of companies building frontier models, the cost of state ownership will not be eliminated.

Governments are rarely the first to build cutting-edge technology or the most creative. Bureaucracies optimize prudence and accountability rather than rapid, haphazard experimentation that produces breakthroughs. Nationalizing the laboratories that build frontier models may give us safer AI, but it is also very likely that we will also get slower and less capable AI.

Why nationalization is not a comprehensive policy

There are layers to the AI ​​ecosystem, and arguments about safety and power apply unevenly to those layers.

At the top sits the foundational model of the GPT, Claude, and Gemini class of systems that Baig and Annett describe that are capable of devastating exploitation. This is the demographic where their claims are strongest. That’s where the nuclear analogy actually has some basis in reality.

Below that lies the infrastructure such as chips, power supplies, and data centers. This is a resource allocation and supply chain issue, not a catastrophic capacity control issue. Taiwan does not need to nationalize TSMC to manage the risk of fraudulent chips. Export controls and standards are required for testing and verification.

And the broadest group includes startups and application developers. Thousands of companies are building narrow, specialized tools on top of models they don’t own or train. This is precisely the demographic where nationalization makes the least sense, as it is precisely the demographic where competition, speed, and trial and error matter most, and where the catastrophic risk argument has little applicability.

Aggregating all three tiers into one policy, namely “nationalization of AI”, is a categorical error. The discussion should really only be about the top layer, and even there it should be done with caution.

incident in singapore

For a country like Singapore, nationalization makes even less sense.

Singapore’s AI strategy was not about building the next frontier model. It’s about applications, tools built by often small private companies that solve specific problems such as logistics, finance, medicine, language, etc.

The country’s economic model is built on foreign investment, free competition, and a reputation for being an easy and predictable place to build a business.

A state-led AI industry will directly compete with all three.

It would signal to foreign investors that the state intends to control the sector in which they plan to invest. This will narrow competition in the very layer where competition improves quality: applications. And since the safety rationale for nationalization has always been about frontier models, not about delivery apps bolting chatbots into customer service lines, it is very likely that a worse product will be produced, not a safer one.

Singapore’s interest is not in owning AI. It’s about ensuring that AI, which is mostly owned and built elsewhere, is properly managed as it emerges.

Better role: referee and sometimes coach

Even if full ownership is the wrong tool for almost the entire AI economy, that doesn’t mean governments can’t do anything. So instead of being the protagonists, they are given the more difficult job of designing the rules of the game and helping the game move quickly in the right direction.

The first lever is the guardrail around the sensitive sector. National security, financial, and critical public utilities are exposed to systemic risks and must be subject to strict and specific regulation.

The second is incentive design. Tax breaks and subsidies can be effective tools that governments can use to encourage AI development toward job-creating tools rather than job-killing tools. Delay the introduction of AI in primary education to avoid slowing down children’s cognitive development. and reward innovation aimed at the public good, in areas such as public health, urban planning, and disaster response, rather than purely private gain.

Third, genuine governance standards. This is where regulation comes into play. Just as new drugs are approved before they go to market, high-risk AI systems will be required to be tested and certified before they go live. Accountability requirements for decisions that significantly impact people’s lives, such as health, credit, employment, and insurance. Conduct independent audits before deployment, not after failure. Mandatory incident reporting when AI systems cause harm. Provenance standards – both for the data used to train models and the labeling of AI-generated content to leave a trail on misinformation.

Singapore has already started building some of that through AI Verify, the world’s first government-developed AI testing toolkit that performs technical audits for fairness, robustness and explainability.

The fourth lever is less discussed, but it deserves equal claim. In other words, the government is not just a referee, but a facilitator.

Singapore already has a working template for this in another industry. The Monetary Authority of Singapore’s regulatory sandbox allows fintech companies to try out new financial products in a controlled environment with relaxed rules and strict supervision before they are distributed to the wider market.

The same logic applies neatly to AI. Sandboxes allow you to test your healthcare AI tools, self-driving car systems, or new financial advisory bots against real-world conditions in a supervised manner, without strangling them with preemptive regulations or releasing them to the public untested.

When done well, sandboxes can do what nationalization never can: accelerate rather than slow down safe innovation.

Baig and Annett are right that AI is no ordinary technology, and states cannot afford to sit idly by while a handful of companies build systems that can harm the entire organization. But the nuclear analogy that makes their argument so compelling also limits it. Uranium can be fenced, but models and applications cannot.

Full nationalization is the wrong move for almost the entire AI economy, and is clearly detrimental to an economy like Singapore built on the private, competitive, application-driven layer of the stack.

What the state can and must do is hold the keys to the few systems that are likely to cause meltdown, judge everyone else with real standards and real teeth, and, through tools like sandboxes, help good ideas advance faster than bad ones. Ownership was never the thing. strait era