For years, the debate around artificial intelligence centered on what the technology could do. Now a more urgent question has emerged: who decides what AI is allowed to do — and what happens when those decisions put tech companies on a collision course with governments?

A Dispute That Made It to Court

A recent report from the United States illustrates just how high the stakes have become. According to published accounts, a federal judge ruled that the Trump administration had illegally retaliated against Anthropic — the AI safety company behind Claude — in connection with a dispute over military use of AI. The full legal details of the case remain complex and not entirely public, but at its core the conflict reflects a fundamental disagreement: the government's desire to deploy advanced AI in defense contexts versus an AI company's effort to control how its technology is used.

This is not a narrow legal matter. It is a signal that the question of who controls AI — and under what conditions — has moved beyond boardroom ethics discussions and into the courtroom. That shift matters.

The tensions are not new. When Google participated in the U.S. Department of Defense's Project Maven — a program using AI to analyze drone footage — thousands of employees signed a protest letter, and Google eventually declined to renew the contract. Microsoft employees raised similar objections to contracts supplying mixed-reality headsets to the U.S. Army. In those cases, internal dissent shaped corporate decisions. The reported Anthropic case appears different in character: an instance where the government itself allegedly took retaliatory action, and a court found that action unlawful.

Why AI Companies Have Usage Policies

Most major AI companies publish what they call Acceptable Use Policies or Usage Policies. These documents define what their AI may and may not be used for. Military applications, autonomous weapons systems, and large-scale surveillance are categories that most leading AI companies explicitly restrict or prohibit outright. Understanding why they do this is worth the time.

The first concern is accountability. When AI plays a role in military decision-making, the question of who bears responsibility for outcomes becomes genuinely difficult to answer. If an AI-assisted decision leads to civilian harm, is the liability with the company that built the model, the government that deployed it, or the soldier who acted on its output? These questions remain legally and ethically unresolved. AI companies, understandably, prefer not to find themselves at the center of that ambiguity.

The second concern is dual-use risk. AI is, at its core, a general-purpose technology. The same capabilities that allow a language model to draft documents, write code, and analyze information can in theory be redirected toward other ends. A model trained or fine-tuned for military use could find its way into civilian applications in ways that are difficult to anticipate or contain.

The third concern is trust — the commercial kind, but also the social kind. AI companies depend on broad public confidence to sustain their business. Being visibly associated with weapons development or battlefield decision-making can damage that trust with the users and enterprise customers who form the foundation of their commercial model. It also shapes how regulators and the broader public think about AI development as a whole.

Where Governments and Companies Clash

Governments, predictably, see things differently. From a national security standpoint, the most advanced AI available should be accessible to defense institutions. The argument is that withholding cutting-edge AI capabilities from the military creates a strategic disadvantage relative to adversaries who face no such internal constraints.

This is not a hypothetical concern. Multiple nations are actively investing in military AI — autonomous drones, AI-assisted logistics, battlefield intelligence analysis, and offensive cyber capabilities among them. The United States, China, Russia, and others are competing in this space. In that geopolitical context, the idea that a private company's internal policy can constrain national defense strategy strikes some government officials as simply untenable.

The conflict takes a particular shape. AI companies set usage policies to define the boundaries of how their technology can be deployed. Governments push against those boundaries, citing security imperatives. When the pushback turns into what a court describes as illegal retaliation, the conflict is no longer just a policy disagreement — it becomes a legal and constitutional question about the limits of government power over private technology companies.

What makes this especially complicated is that there is no neutral ground. AI is powerful enough that both sides have legitimate stakes. The risks of misuse are real. So are the risks of one nation falling behind another in capabilities that may determine outcomes in future conflicts. Neither concern goes away simply by asserting the other.

What This Means for AI Governance

The deeper issue this dispute raises is a governance gap. Currently, the usage policies created by private AI companies serve as a de facto rulebook for what the technology may and may not be used for. There is no binding international treaty governing military AI. Domestic regulations vary widely and remain incomplete. The frameworks that do exist are mostly voluntary or aspirational.

The European Union has begun to close part of this gap through its AI Act, which targets high-risk AI applications and establishes compliance requirements across sectors. The United Nations has ongoing discussions about lethal autonomous weapons systems. But enforceable international norms with real consequences for violations are still in early formation. The governance infrastructure has not caught up with the technology.

In the absence of a clear legal and international framework, conflicts like this one are likely to multiply. Governments will continue to push for access to frontier AI for security purposes. AI companies will face pressure — competitive, commercial, and increasingly legal — to either comply or resist. Courts may find themselves adjudicating disputes that were once settled behind closed doors, without the precedents or regulatory guidance needed to resolve them cleanly.

The question of who controls how AI is used is not, in the end, a technical question. It is a political, ethical, and legal one. The technology itself is neutral — it does the work it is directed to do. The answers to questions about that direction will be worked out in legislatures, courtrooms, international forums, and the internal policies of companies that have built systems far more capable than most anticipated even a few years ago. This debate is only beginning, and watching how it develops is worthwhile regardless of where you stand on the underlying questions.