Pentagon looks for an Anthropic replacement after clash over limits on military use of Claude

The Pentagon has begun preparing to replace Anthropic's tools after a sharp conflict over how the US military can use commercial AI models. The company refused to lift some of the restrictions on Claude, and the administration of Donald Trump responded by designating it a "supply chain risk."

Why the relationship broke down

The dispute is not about model quality or contract price. The disagreement is over who ultimately sets the rules for using AI in the military system. According to a senior US Department of Defense official, the Pentagon insisted on the right to use the technology "for any lawful purpose." Anthropic, by contrast, wanted to write two hard limits into the agreement: a ban on mass surveillance of Americans and a ban on fully autonomous weapons without human involvement.

The conflict quickly went beyond ordinary contractor negotiations. After the sides failed to agree, the US administration gave Anthropic the supply-chain risk status, a label usually applied to threats to government procurement and critical infrastructure. For the company this is not just reputational damage: the status can complicate work with military contractors and federal agencies, even where no direct contract with the Pentagon is involved. A dispute over guardrails turned into a dispute over access to the government market within a few days.

How the Pentagon is looking for a replacement

The military did not wait for the dispute to end and began building a fallback plan. According to a Pentagon representative, the department is already doing engineering work to deploy several large language models in government-controlled environments. The idea is to avoid depending on a single supplier and to get alternatives for operational use as quickly as possible.

This is not a long-term study but preparation of working solutions that can be built into sensitive processes fairly quickly. Judging by the department's statements, the Pentagon wants to replace not only a specific product but the very model of dependence on a single AI partner. That means a broader stack of suppliers, tighter control over the deployment environment, and a faster move to compatible solutions if relations with an individual vendor reach a dead end.

• moving AI tools into government environments, not only the vendor's infrastructure
• connecting several LLMs instead of relying on one model
• rapidly putting alternatives into service for sensitive scenarios
• reducing Anthropic's role in existing military workflows

Against this backdrop, it is telling that the market is quickly filling the vacuum. OpenAI has already agreed with the Pentagon to run its models in closed environments, and the military shows it is ready to widen its pool of suppliers if one of them tries to define firm limits on acceptable use. For the state, this is a question of operational resilience. For AI companies, it is a signal that defense contracts will increasingly demand not only technology but also a willingness to accept the customer's terms.

Anthropic's position

Anthropic did not soften its tone even after the Pentagon's sanction and made clear it considers the authorities' actions legally questionable. Company chief Dario Amodei said Anthropic has "no other choice" but to challenge the decision in court. For the company this is a matter of principle: it is trying to defend an AI supplier's right to refuse scenarios it considers dangerous, even within formally lawful use. In other words, the dispute is not only about a contract but also about a business's right to keep its own red lines.

"We do not consider this decision legally sound and see no other

choice but to challenge it in court."

The Pentagon's position is the exact opposite. The department holds that a supplier cannot stand between the military and the lawful use of a critical technology. That makes the current dispute look like an early test of the future balance of power: whether large AI companies can impose their own guardrails on the state, or whether the state will insist on its rules in full. That is why a conflict over a single model has suddenly become a precedent for the entire defense AI industry.

What this means

The Anthropic episode shows that the main dispute over military AI right now is not about which model is smarter but about who controls the limits of its use. If the Pentagon really does replace Claude quickly with alternatives, it will be a harsh signal to the whole market: refusing to lift restrictions can cost access to the largest government contracts. At the same time, it pushes AI companies to decide in advance where they draw the line between safety, politics, and revenue.