The Download: The Pentagon’s new AI plans, and next-gen nuclear reactors

The Pentagon is preparing for a change that could fundamentally transform how the U.S. military uses artificial intelligence. Rather than waiting for ready-made solutions from OpenAI or Anthropic, the military plans to create secure environments where AI companies can train models on classified data. This is not a typical collaboration — it is a strategic move that opens the door to the militarization of generative artificial intelligence on a scale we have only seen in science fiction films.

The Pentagon's initiative stems from a simple observation: the largest AI models are trained on public data, which means they cannot be optimally tailored to special defense needs. Defense ministry officials argue that access to classified information would allow AI to better understand military contexts, from battlefield analysis to operational planning. This sounds rational — until we start thinking about potential threats.

In parallel with these plans, the energy sector is undergoing its own revolution. Next-generation nuclear reactors promise to change the planet's energy landscape. These are no longer the bulky, cumbersome devices from the seventies — they are compact, modular systems that can power cities or factories without occupying hectares of land. But will this technology come to Poland, or will it remain a Western privilege?

The Pentagon and artificial intelligence: a strategy without borders

The U.S. Department of Defense has been grappling with a challenge for years: how to integrate rapidly developing AI technology with security and classification requirements? The traditional approach — purchasing licenses for existing models — was not sufficient. ChatGPT does not know the specifics of electronic warfare. Claude does not understand drone tactics. This is precisely why the Pentagon decided to take a step further.

The project envisions the creation of special servers and facilities where employees of AI companies can work with classified data without the ability to export or copy it. The scenario is straightforward: an engineer from OpenAI travels to a military base, works in an isolated environment, and trains a model on data about defense systems, strategies, or threat analysis. The model remains in the Pentagon's possession, but the technology and know-how come from civilian tech giants.

This solution has elegant logic, but it raises serious questions. First — what will happen if the model is leaked or intercepted? Second — do AI companies really want to be this close to military operations? OpenAI and Anthropic have already expressed concerns about military applications of their models. Will they change their minds under pressure from government contracts worth billions of dollars?

The Polish perspective on this issue is particularly interesting. Poland does not have access to such programs — at least not officially. Our military must rely on solutions available on the open market or on allied NATO technologies. This means that in the race for AI military advantage, we are at the back of the pack. Is this a problem? Definitely, if we consider that AI capabilities could be decisive in future conflicts.

Next-generation nuclear reactors: the energy of the future or a mirage?

While the Pentagon plans its digital future, the energy sector faces its own transformation. Small Modular Reactors (SMR) are not just a change in size — they are a change in the philosophy of energy production. Instead of one gigantic reactor costing billions of dollars, which must be built over ten years, we can have factories producing modular reactors that can be assembled like Lego blocks.

SMR technology has real advantages. First, these reactors can be located closer to energy consumption points, eliminating transmission losses. Second, they are safer — their smaller size means less fuel and less energy to disperse in case of an accident. Third, they can power heavy industry that cannot wait for variable solar or wind energy. Steel mills, refineries, cement factories — all these sectors can finally decarbonize without relying on batteries or energy storage.

But there is a catch. Next-generation reactors are still much more expensive per megawatt than traditional energy sources. Capital costs are enormous, and return on investment takes decades. Additionally, the regulatory infrastructure of most countries is not ready. In Europe, the approval procedures for a new reactor can take as long as its construction. In the USA, the situation is somewhat better, but still complicated.

Poland has an opportunity here that it should seriously consider. Our energy plans assume the construction of several large nuclear reactors — this is the right decision. But should we not also invest in SMR? These reactors could power developing cities, industry, or even individual large facilities. The Ostrołęka power plant, instead of being coal-fired, could be powered by small nuclear reactors. That would be a much more forward-thinking solution than gas or coal.

Data security vs. national security

Let us return to the Pentagon and AI. The defense ministry's initiative reveals a deep tension in the modern technological world: data security and national security often collide with each other. AI companies build their models on the principle of openness — the more data, the better the model. But classified data by nature must be closed.

The Pentagon attempts to solve this paradox through physical isolation. An OpenAI employee works in a secure room, the model trains on classified data, and then the model remains in the Pentagon's possession. But this solution has scaling limits. How many such facilities can the Pentagon maintain? How many employees from OpenAI, Google, or Anthropic will want to work in military bases?

There is another layer of complexity. AI models trained on military data may unintentionally store sensitive information in their neural weights. Research shows that large language models can reproduce fragments of training data if asked appropriately. Will the Pentagon be able to prevent information leaks through model extraction? That is unknown.

For Polish technology companies, this presents both a challenge and an opportunity. A challenge, because we will be even more dependent on technology developed in the USA. An opportunity, because the Polish cybersecurity industry could develop solutions to monitor data leaks from AI models. This is an area where Poland has genuine expertise.

Nuclear energy in Poland: from theory to practice

Polish energy is at a crossroads. The European Union is pushing for decarbonization, and we must say goodbye to coal. Nuclear power plants are an obvious choice — they are low-emission, reliable, and can power the base load of the grid. But large reactors are more expensive and slower to build. Small reactors could be an interim or complementary solution.

Polish Atomic Energy, the company tasked with managing the nuclear program, should consider including SMR in its strategy. Not instead of large reactors, but alongside them. Modular reactors could power:

• Heavy industry — steel mills, refineries, chemical factories requiring process heat
• Developing technology centers — large server farms, data processing centers
• Regional cities — as a reliable energy source to complement renewables
• Water desalination — in the future, if we face fresh water shortages

Costs are indeed a challenge. But if we look at the total calculation — investment costs, operating costs, CO2 emission costs, reliability costs — small reactors can be competitive. Especially if the technology develops and scales, as all new technologies do.

Military AI vs. commercial AI: diverging paths

The Pentagon wants military AI, but that does not mean commercial AI will develop in the same direction. In fact, the militarization of AI could lead to fragmentation of the technological ecosystem. We will have models trained on public data (OpenAI, Google, Meta) and models trained on military data (Pentagon). These two worlds may increasingly diverge.

For startups and technology companies, this could be both a threat and an opportunity. A threat, because access to military data and funding will be limited to selected companies. An opportunity, because new niches may emerge — companies specializing in AI for specific sectors (medicine, law, education) can develop independently of military ambitions.

The Polish technology industry should be cautious. We do not want to be dependent on military contracts from the West, but we also do not want to be completely left out. The strategy could be to build independent AI capabilities, particularly in areas that are important for the Polish economy and security.

Nuclear reactors and energy security

Small modular reactors are particularly interesting from an energy security perspective. Poland, located between Germany and Russia, must be aware of its energy vulnerabilities. If we have distributed, modular sources of nuclear energy, we will be less vulnerable to supply disruptions or infrastructure attacks.

This is not speculation — it is a real strategy being considered by many countries. Sweden, for example, is considering expanding nuclear capacity, including SMR. Germany, despite its obsession with renewable energy, is also looking at nuclear energy as a necessity. Even France, which already has a lot of nuclear energy, is considering investments in SMR.

Poland should hurry. If we wait too long, SMR technology will already be deployed in other countries, and we will have to buy licenses instead of building our own expertise. This was a mistake Poland already made with renewable energy — now we buy solar panels instead of producing them.

The future: fragmentation or integration?

The Pentagon is planning military AI, and the energy industry is planning a nuclear revolution. These are two processes that will shape the future for decades to come. But the question is: will these be processes in which Poland is involved, or will we observe them from the sidelines?

On the issue of AI, Poland's position is weak. We do not have access to Pentagon military programs, and our technology companies are too small to compete with OpenAI or Google. But we can develop niche AI applications — for Polish business, for the Polish government, for Polish security. This requires investment in education, research, and infrastructure.

On the issue of nuclear energy, we have a chance. Poland already has experience with atomic energy (albeit minimal), and the energy market in Poland is large. If we act quickly, we can become a hub for SMR technology in Central and Eastern Europe. This would be significant for our economy and security.

Ultimately, both the Pentagon's military AI and next-generation nuclear reactors are symptoms of the same phenomenon: technology changes faster than policy and regulations. The Pentagon is trying to catch up with change by controlling access to data. The energy industry is trying to catch up through innovative projects. Poland should do the same — not wait for other countries to chart our course, but actively shape the future.