US Justice Department: Anthropic unlawfully tried to restrict military use of Claude

The US Justice Department says it lawfully penalized Anthropic for trying to restrict military use of its AI models. In doing so, it signaled that private companies' ethical restrictions cannot dictate the terms of government contracts.

The conflict surfaced when Anthropic sued the US government. The company claims the authorities unlawfully punished it for trying to control how the military uses Claude models. In its response, the Justice Department took the opposite position: the government acted within the law, and Anthropic tried to unilaterally limit how technology supplied under government contracts could be used.

At the heart of the dispute is Anthropic's acceptable use policy, which the company aims to apply to every customer without exception. Anthropic openly presents itself as one of the most responsible AI developers. Its safety principles prohibit using Claude to develop autonomous weapons systems, for targeting, or for other combat tasks without adequate human oversight. In Anthropic's view, these restrictions apply to everyone, including government agencies.

The government flatly disagrees. Its position is that when it buys AI technology, it gains the right to use that technology according to its own operational needs. It treats Anthropic's attempts to limit that right not as a legitimate exercise of ethical principles but as a breach of contract terms. That, the Justice Department argues, is what justified the sanctions against the company.

The case reaches far beyond a single legal dispute. The US military and intelligence sectors are actively seeking partners among AI developers, and they keep running into the fact that the industry's leading players have their own views on acceptable uses of their technology. Microsoft, Google, and Palantir have already signed major contracts with the Pentagon, effectively without imposing similar restrictions. Anthropic, apparently, was trying to build a different model: keep the government contract while retaining control over how Claude is used.

The government's stance in this dispute matters for the entire industry. If the court sides with the Justice Department, it will set a precedent: AI companies working on government contracts will not be able to restrict military use of their products through user agreements. If Anthropic wins, developers will gain a mechanism for building ethical restrictions directly into their contract terms with the government.

More than money is at stake. Anthropic raised billions of dollars in investment largely on the argument that safety is a real competitive advantage for the company, not a marketing slogan. If it backs down on military use of Claude without a legal fight, the reputational damage could outweigh the financial losses. Seen this way, the lawsuit reads as an attempt to lock in its principles legally.

The conflict is part of a broader debate: to what extent are AI companies responsible for how their products are ultimately used? Some experts argue that ethical restrictions built into a product cannot, by definition, be binding on a government customer. Others insist that such mechanisms are exactly what preserves a minimum level of control at a time when AI systems are becoming ever more deeply embedded in military decision-making.