Spain's UN AI Expert Warns of Tech Power Concentration: 'We're Just Scratching the Surface'

Román Orús sits on a committee of forty. The United Nations asked him to help figure out what to do with artificial intelligence, and he is the only Spaniard among them. He is a physicist at the Donostia International Physics Center in the Basque Country, runs a quantum computing company, and now finds himself at the center of a question that may define the next decade: who gets to control the most powerful technology humanity has ever built?

AI arrived as a promise. Machines that could learn, heal, invest, reverse climate change. Instead, it has entangled itself in wars, economies, elections. The question of control is no longer theoretical. It is urgent. Orús does not mince words about what happens if the answer goes wrong. "There is no alternative to preventing it," he says, referring to the concentration of AI power in the hands of a few corporations. "The alternative to preventing it is zero."

What makes his warning distinct is his refusal to treat this as a simple problem with a simple solution. AI is not like the atomic bomb, he explains, because the atomic bomb was built by governments. The first large language models were built by private companies. The scientific papers describing them are published. Many are open source. It is, he says, like putting doors on an open field. You cannot really lock it. Yet something must be done, because the alternative is a world drowning in disinformation, where nothing seen on social media can be trusted, where cybersecurity risks multiply faster than defenses can be built. Anthropic had to shut down one of its own models because it was too powerful, too capable of causing harm. This will only accelerate.

Pedro Sánchez, Spain's prime minister, called it "silent colonialism" when he opened the UN panel in Madrid this spring—the quiet transfer of power from elected governments to private technology companies. Orús acknowledges the reality without flinching. The politicians make the rules, he says, but the companies holding the pan are the ones building the models. They are the ones who must adapt to whatever regulation emerges. Yet he sees something larger at work: a society drifting toward technocracy, where power accumulates not in the hands of democratic representatives but in the hands of private capital. This is bigger than AI. It is a question about what kind of capitalism survives the twenty-first century.

The UN committee will not simply hand a code of conduct to the tech companies and ask if they agree. That is not how this works. They will produce reports—the first in July—and they will keep producing them because the field moves too fast to stand still. They are a hinge, Orús says, between the scientific world and the politicians who actually regulate. The lesson from social media is already written: a technology expands globally before anyone understands its consequences. Mark Zuckerberg did not invent Facebook to destroy democracy. He wanted to connect with friends. The same is true of language models, of quantum computing, of biotechnology. Every technology since humans discovered fire has carried both opportunity and risk. The wheel moves carts and builds tanks. AI will do both. The question is whether we develop it well.

On the geopolitical plane, the stakes multiply. AI is no longer just a tool. It is a weapon. Information warfare is the dominant form of conflict now, though it translates to the battlefield. Palantir, the data company, has become a symbol of how governments outsource military functions to private firms—a surrender of sovereignty dressed in a contract. When the United States bombed Iran, Iran responded by attacking Amazon data centers in the Gulf, cutting off taxi services and internet across the region. Tech infrastructure is now a legitimate military target.

But there is a deeper problem that touches daily life more directly: the algorithms that make decisions about us are black boxes, even to the engineers who built them. No one fully understands why an AI model decides what it decides. This is not just a problem for weapons systems or financial models where a client needs to know why they should invest. It is a problem for everyone. Even the technicians do not know exactly what happens inside the machine when it chooses one path over another. When AI hallucinates, giving strange answers to straightforward questions, it is because it is identifying patterns and making predictions based on correlations humans cannot see. When it predicts poorly, it invents. That is all.

The future architecture of AI is still being written. Orús believes it will not be a single centralized model controlling everything. Sam Altman at OpenAI once thought it would be. He has changed his mind. Instead, there will be a spectrum: enormous models like ChatGPT for complex tasks, but also smaller intelligences embedded in phones, tablets, watches, cars, drones, satellites—specialized systems running on limited hardware, smart enough to do one thing well. These agents will interact with each other, coordinating automatically, doing work without human intervention. Some models can already write presentations, build spreadsheets, order food. This is the direction. It is not without problems. Energy consumption is one. How these systems coordinate with each other is another. The alignment has to be correct, and at some point a human has to check that nothing strange is happening. There is an entire story unfolding, and we are only beginning to see its shape.