Experts Share: Who Should Control the World’s Leading AI Models?

Governments, tech companies, academics and civil society are meeting at the United Nations Global Dialogue on AI Governance in Geneva to discuss a very important question: who should control AI?

The meeting came after the publication of the first report from the United Nations Independent International Scientific Panel on Artificial Intelligence, which is a group of 40 experts from every region of the world.

Many people agree AI can help economies, healthcare and science. The harder question is who gets to write the rules, especially as much of the world’s most advanced AI comes from just the United States and China. We know that recently, OpenAI proposed that the US government should take a 5% stake in the company, which raised a lot of questions on what the intention here could actually be.

What Happens If Only A Few Countries Make The Rules?

Ambassador Rein Tammsaar of Estonia said, “For many countries in the world, AI could be a great equaliser. It can support economic development, advance competitiveness, support science and health systems. Machine learning in general could benefit productivity. This is the potential.”

He also said, “The frontier developers are basically concentrated in two countries [US and China]. This leaves other countries with a lot of questions.”

Ambassador Egriselda López of El Salvador said, “The AI divide is real. Some countries have very strong infrastructure and strong skills and research capacities. Whereas there are others that are still struggling with issues like connectivity and public infrastructure.”

López also said, “The Global Dialogue is the first platform in the United Nations for the discussion of AI governance. It’s also an opportunity for Member States to come together to have an inclusive discussion; But not only governments, it’s also about bringing together different stakeholders.”

Can Anyone Keep Up With AI?

Professor Yoshua Bengio, co chair of the Scientific Panel, said, “AI is approaching or surpassing human capabilities in many domains. It is outpacing both scientific understanding and governments’ ability to adapt. There have been incredible advances which are changing the world and it doesn’t look like it’s stopping.”

He added, “With growing evidence of deceptive AI behaviour, science currently cannot guarantee that as capabilities continue to increase, AI will not cause catastrophic harm, either on its own or due to malicious users.”

Maria Ressa said, “The world cannot govern what it cannot understand. The Panel’s report provides independent science, drawn from every region, and available to every government. Its message is clear: the potential is great, but the risks are real, and the cost of waiting is rising.”

Ressa also said, “Not one country can actually deal with this technology on its own; It needs to be a multilateral solution. And the body that is set up that could do this is the United Nations. Now the question is, will its Member States move?”

Ranjith Raghunath, CEO of CX Data Labs, says, “There just isn’t an international body with the power, expertise, and respect to take on this task, especially in a multipolar world where countries see AI as potential tools for warfare, cyber and otherwise. Due to its potential impacts and how resource intensive it is to deploy, some kind of national or international oversight would be a good idea, but I simply don’t see it happening any time soon.”

More experts answer the question, “who should control the world’s AI?” and here’s what they say…

Our Experts:

• Roi Carmel, CEO and Co-Founder, Spotlight.ai
• Olga Kokhan, CEO, Tinkogroup
• Hien Nguyen, Founder and CEO, Screate Labs
• Lily Dash, Founder, Future Caribbean
• James Cranwell, Head of Product, 5app
• Kate O’Neill, Founder and CEO, Tech Humanist, Global Keynote Speaker and Author, KO Insights
• Adam Dalloul, Founder, EmpirioLabs AI
• Matt Baharav, Founder and Chief Executive Officer, MKB Media Solutions

Roi Carmel, CEO and Co-Founder, Spotlight.ai

“Who should control the models” is the wrong question, because model-level control is nearly impossible to enforce and it’s not where the harm happens. A model sitting on a server harms no one. The damage and the value both show up at the point where AI makes a decision that affects a real person: a loan, a diagnosis, a hire, a forecast someone bets a business on. That’s where accountability has to sit, and it can’t sit with a global committee that meets twice a year.

“I build autonomous AI for a living. I see every day how confidently these systems produce wrong answers when you hand them a big question and partial data. My whole thesis is to let humans excel at being human and let AI quietly handle the thousands of signals and decisions that get in their way, and the only way that works is guardrails at the point of use.

“The organisations deploying AI into real decisions understand its failure modes, and they’re the ones positioned to govern it in the moment. Attach governance to deployment, with clear liability for whoever puts the system in front of a decision. One body controlling the frontier sounds orderly and does nothing about the model already running inside ten thousand companies tonight. Regulate the use, name the accountable party, and stop pretending the leverage is at the top.”

Olga Kokhan, CEO, Tinkogroup

“I don’t think the world’s most advanced AI models should be in the control of any one government, company, or international organisation. AI is too powerful to be controlled by a single group. Instead, I think we need a shared governance model involving governments, researchers, technology companies, independent experts, and civil society. They all have different views about innovation, security, ethics, and public impact, and we need all those views.

“The global guardrails should focus on areas where broad agreement is possible, including transparency, safety testing, accountability, and preventing clearly harmful misuse. At the same time, regulation should not become so rigid as to stifle responsible innovation or erect barriers that only the largest companies can overcome.

“As the founder of Tinkogroup, a data services company that helps develop AI through data annotation and processing, I’ve noticed that building trustworthy AI is as much about responsible processes and human oversight as it is about the models. Effective governance should promote innovation and accountability, not force them to compete.”

Hien Nguyen, Founder and CEO, Screate Labs

“I think we’re asking the wrong question.

“The debate shouldn’t be “Who should control the world’s leading AI models?” It should be “What should anyone be allowed to control?”

“Control is a dangerous framing because it assumes AI needs a global owner. Most foundational technologies don’t succeed because they’re controlled by a single institution. They succeed because they operate on shared protocols and common rules.

“For example, no single institution controls the internet.

“Instead, we govern the incentives around it: safety, liability, interoperability, competition, and abuse. And the internet wasn’t built around a global controller, it was built around protocols. TCP/IP. HTTP.

“AI should be no different.

“The real challenge isn’t deciding who sits at the top. It’s preventing any single government, company, or institution from becoming a permanent gatekeeper for intelligence itself.

“Global governance should focus less on controlling models and more on designing protocols that make AI trustworthy, auditable, interoperable, and competitive. Protocols outlast companies. They outlast governments. That’s what creates resilient systems.

“If we optimise for control, we concentrate power. If we optimise for protocols, we distribute opportunity.”
 

Lily Dash, Founder, Future Caribbean

“No single government, company or international body should have the power to decide how the world’s most consequential AI models are built, deployed or governed, because the technology is already shaping economies, information systems and public life across countries with very different needs, capacities and priorities.

“The role of global institutions should be to set a shared floor around safety, transparency, accountability and human rights, while ensuring that the countries most likely to be affected by AI are not simply asked to adopt rules written elsewhere. Governments need a meaningful role, but so do researchers, civil society, technical experts, founders and communities whose languages, labour markets and public services will be changed by these systems.

“If a small group of large technology companies and wealthy countries control the models, the infrastructure and the policy agenda, then AI will reflect a very narrow view of the world. Through my work with founders building AI products across emerging markets, I see how much talent exists outside the traditional technology centres; governance should protect people from harm while making sure those builders have a real seat at the table, access to the tools and infrastructure they need, and a fair chance to shape what comes next.”

James Cranwell, Head of Product, 5app

“No single entity should have control over the world’s leading AI models. This sets a dangerous precedent, as anyone with control of the most powerful models will have a huge advantage on the world stage – especially as we move towards the age of Artificial General Intelligence (AGI), which will quickly become a significant strategic resource alongside land, energy, finance and military power. If one entity has control over AI, powerful governments and corporations could use these systems to manipulate markets, security and the media, leading to profound political divides.

“Going forward, governments and AI-focused businesses must work together to create a set of established guardrails. The EU AI Act in Europe is a good starting point, as it prioritises human users of AI tools to ensure they stay fully informed and protected. Human control, human dignity and democratic accountability must sit at the heart of any decision made about the use of AI, and any entity responsible for AI tools must consider transparency, user privacy, government regulation in the public interest and equitable access to AI tools and solutions. If AI control is limited to a single entity, there’s a risk that the most powerful AI tools could be restricted to the wealthiest people in society, which should never be the case.”

Kate O’Neill, Founder and CEO, Tech Humanist, Global Keynote Speaker and Author at KO Insights

“Your question is well-timed (as I’m sure you know). As I write this comment back to you, the UN’s first Global Dialogue on AI Governance is wrapping up in Geneva, wrestling with exactly this.

“Where I land on the question is this:

“No single entity should control the world’s most powerful AI models, whether that’s one company, one government, or one global institution. Putting that much power in any single place is its own risk, and it’s the failure mode we should be working hardest to avoid.

“The workable answer is layered: broad international agreement on a few human-centered non-negotiables, and democratic institutions with real authority to enforce them close to the people affected.

“We already have a template for this in human rights. Countries disagree about almost everything, yet many can still agree that certain uses cross a line: mass biometric surveillance, weapons that kill without human judgment, systems built to resist shutdown, manipulation dressed up as personalisation, high-stakes decisions with no human accountability or recourse.

“Naming those limits is work for international forums like the UN. Enforcing them falls to member states, courts, regulators, and laws written close enough to voters to matter, which is why the EU AI Act and Illinois’ biometric privacy law are worth watching. Principles are the easy part. The harder, more telling work is enforcement: whether these limits hold up against money, procurement pressure, and geopolitical competition, or fold the moment they become inconvenient.”

Adam Dalloul, Founder, EmpirioLabs AI

“I don’t think it should sit with any single authority. The risk isn’t really lowered by that kind of restriction anyway. If someone has a goal for mal-intent usage of AI, there are ways around this if people are dedicated enough. There are a variety of open-weight models that are now catching up to the abilities of SOTA proprietary models from the big US labs. It’s only a matter of time until that gap is bridged, and GLM 5.2 is a notable example here.

“So whoever ends up setting the rules, the capability is already spreading out on its own, and heavy centralised control mostly slows down the people who follow them.”

Matt Baharav, Founder and Chief Executive Officer, MKB Media Solutions

“Right now, there is tremendous pressure to centralise control over these new models.

“Every major government wants to be able to choke the throat of that one person responsible when something goes wrong with that model.

“But giving total control to a global committee or a few tech giant companies will completely strangle independent innovation.

“Right now, there is a narrative out there saying we need massive, unified guardrails to keep the peace.

“The reality is that creating guardrails and centralising control will create an enormous bottleneck that only the richest corporations can afford to bypass.

“We do not need a global registry for approved models.

“What we need are open standards focused on transparency & data privacy so individual developers can build without having to ask permission from a bureaucratic authority.”