Nobody knows how AI companies should work with the government

There is a particular irony in the fact that the company that built the world's most popular AI system turned out to be wholly unprepared for the consequences of its own success. OpenAI began as a nonprofit research lab with an idealistic mission, then became one of the most valuable startups on the planet. It is now entering the third phase of its evolution: becoming a component of United States national security. And, as TechCrunch rightly notes, the company has no tools, no experience, and not even a clear roadmap for this role.

The problem, however, runs far deeper than the difficulties of a single corporation. No country in the world has yet articulated a coherent model for how government institutions should work with private AI developers. Historical analogies work poorly here. When the US government built its relationships with defense contractors such as Lockheed Martin or Raytheon in the mid-twentieth century, these were companies created from the start to work with the state, fluent in the language of bureaucracy and ready for strict regulation. AI labs are a different story altogether. They grew out of academia and Silicon Valley venture culture, where speed matters more than procedure and openness matters more than secrecy.

OpenAI's transformation is especially telling. Sam Altman's company has covered in two years a path that took traditional tech giants decades. ChatGPT became the fastest-growing consumer product in history. GPT-series models are used by millions of people and thousands of companies worldwide. But alongside the commercial success grew the recognition that large language models are not just a handy tool for writing emails and generating code. They are a technology with obvious dual use, capable of influencing the information space, cybersecurity and, in time, the military balance of power.

American officials understand this. The US intelligence community is already experimenting with closed versions of language models for data analysis. The Department of Defense is studying how generative AI could be applied to logistics and planning. But every such project runs into a fundamental question: on what terms should a private company, answerable to its investors and users, provide its technology to government bodies answerable to entirely different priorities? Who defines the limits of what is acceptable? Who controls how the models are actually used once handed over?

Despite all its rhetoric about responsible AI development, OpenAI has so far not demonstrated convincing mechanisms for managing these risks. The company's internal processes remain opaque. Its board, which in theory should balance commercial interests against the public good, went through a dramatic crisis in late 2023 and has since been rebuilt into a configuration considerably more loyal to Altman. The shift from a nonprofit structure to a commercial one, which the company is completing in 2025–2026, only heightens concerns: who exactly will make sure that technologies capable of changing the global balance of power are used responsibly?

For Russia, this story carries a double significance. On one hand, the lack of a clear model for relations between AI companies and the state is not a uniquely American problem. Russian developers, from Sber to Yandex, also balance commercial goals against state interests, albeit in a very different institutional environment. On the other hand, whichever model the US ultimately chooses will inevitably affect global norms, just as American internet-regulation standards once set the frame for the entire world.

The central paradox is that the time for calmly designing the rules has already passed. The technology is moving faster than institutions can adapt. OpenAI is already a de facto strategic asset. The only question is whether that status will be formalized through a well-thought-out system of mutual obligations or remain a gray zone where decisions are made ad hoc, under pressure of circumstances. For now, all signs point to the second option. And that is perhaps the most worrying news not only for America but for the entire global AI industry.