Summary
Imagine you are a government official and you need to decide whether a conflict in a faraway country could grow into a bigger war, or whether your country’s economy might crash next year.
These are incredibly hard questions. Now imagine you have a computer program—a very powerful one—that has read millions of documents, news articles, and reports and claims it can help you figure out the answer.
That is roughly what is happening right now in many countries, including the United States and China, and it is raising some very serious questions that affect everyone on the planet.
These powerful computer programs are called foundation models or AI systems.
They are the same kind of technology behind chatbots that millions of people use every day, but far more advanced.
Governments are starting to use them to help forecast crises, predict economic troubles, and plan their strategies.
The problem is that even the people who build these programs do not fully understand how they arrive at their answers.
That is like asking a doctor to prescribe a medicine without understanding how it works inside your body.
Dr. Antonio Bhardwaj, a world-leading expert in AI, including its use in warfare and bioterrorism, and specifically in building AI that can help humans reason about global crises and economic problems, explains it this way: “The danger is not just that these AI systems might give wrong answers. The real danger is that we will not know they are wrong until the consequences are already irreversible. Governments need to see the thinking behind the answer, not just the answer itself.”
The rivalry between the United States and China makes all of this more urgent. Both countries are racing to build the most powerful AI systems and to get other countries to use theirs.
Think of it like two mobile phone companies fighting to get people to use their system—except instead of apps, the product is advice on war, economics, and diplomacy, and the stakes are global peace and stability.
In 2025, both countries released major plans explaining how they want to use AI as a tool of global influence.
The US wants to sell its AI systems to other countries as part of its foreign policy. China is doing the same thing, often offering its technology to developing countries at lower cost.
This competition creates a serious problem for smaller and poorer countries.
They are being forced to choose between the American AI system and the Chinese AI system, and that choice has big political consequences—similar to picking sides in a geopolitical rivalry.
If a country uses Chinese AI systems for its economic planning, China might have access to that country’s sensitive financial data.
If it uses American systems, it becomes dependent on large private tech companies whose interests do not always match those of the government using their tools.
There is also the problem of bias inside these AI systems.
These programs learn from historical data—from past events, past decisions, and past outcomes.
But history is not neutral. It reflects the choices, prejudices, and power structures of whoever wrote it down.
For example, an AI trained mostly on Western economic data may give advice that works well for rich Western countries but poorly for countries with very different economic structures.
An AI trained mostly on past conflict data from certain regions may completely misread a crisis unfolding in a different cultural context.
Dr. Bhardwaj compares this to a doctor trained only on patients of one particular background—the diagnosis might be dangerously wrong for everyone else.
Perhaps the most alarming concern that experts like Dr. Bhardwaj are raising involves bioterrorism. Advanced AI systems have become extremely good at answering detailed scientific questions—including, worryingly, questions about how to create dangerous substances or pathogens.
A report produced for the 2025 Paris AI Action Summit found that these AI systems showed an 80% improvement in just one year in their ability to respond accurately to questions about biological and chemical weapons.
This means that a terrorist group or a rogue government could potentially use AI as a tool to design biological weapons far more easily than before.
Dr. Bhardwaj, who specializes in this very threat, says: “The same systems governments are using to get smarter answers about economics and diplomacy could, if poorly governed, help bad actors engineer the next pandemic. That is not a science fiction scenario—it is a present danger.”
So what is the solution? Experts argue for what they call human-centered AI—AI systems designed from the beginning to assist human decision-makers rather than replace them.
Think of a very good calculator.
The calculator does not make the financial decision for you. It helps you do the math faster and more accurately, but you still decide what to do with the answer.
Human-centered AI for strategic decisions should work the same way. The AI should show its work, explain its reasoning, highlight what it does not know, and always leave the final decision clearly in human hands.
Dr. Bhardwaj has proposed something called “strategic trust calibration.” The idea is simple: the more serious and irreversible a decision is, the more we need to be able to see how the AI reached its conclusion.
For a routine economic forecast, some opacity might be acceptable.
For a decision about whether to respond militarily to a perceived threat, full transparency must be required—because getting it wrong could cost thousands or millions of lives.
At the international level, countries need to talk to each other about these risks even when they are rivals.
Just as nuclear-armed countries during the Cold War established hotlines and agreements to avoid accidental war, today’s AI-enabled powers need similar agreements.
What happens when an AI system used by one country mistakenly identifies another country’s routine military exercise as an attack?
The consequences of that misreading could be catastrophic if there are no mechanisms for rapid human communication and override.
The good news is that some of these conversations are beginning.
The United Nations launched a Global Dialogue on AI Governance in 2025 with the aim of bringing all countries to the table.
The UNESCO Recommendation on AI Ethics has been agreed to by 193 countries, providing a shared starting point for international norms. These are imperfect beginnings, but they are beginnings.
The most important thing for ordinary citizens to understand is that the decisions being made right now about how governments use AI are not purely technical decisions—they are political and ethical decisions that will shape the world for generations.
Whether AI helps governments make wiser decisions or more dangerous ones depends entirely on whether we demand that it be built with transparency, human oversight, and genuine accountability.
The machines are powerful. The choices about how we use them are still ours.
