Appeals Court Lets Pentagon Keep Anthropic Out of Its Systems

The Pentagon can keep treating Anthropic as a supply-chain risk after a federal appeals court in Washington, D.C., declined to overturn one of the government’s restrictions on the AI company. The ruling matters because it preserves a military block on Claude, Anthropic’s flagship model, even as the company moves toward a possible public offering and competes for enterprise and government business.

In a 2-1 decision on Friday, the U.S. Court of Appeals for the D.C. Circuit said the Defense Department had enough basis to conclude that keeping Claude integrated into military information systems posed a national-security risk. The court also rejected Anthropic’s arguments that the government violated its due process and free-speech rights, leaving the Pentagon’s position intact for now.

The decision adds another layer to a widening legal and policy fight over whether AI companies can restrict how their models are used by the U.S. government, especially for military and surveillance purposes. It also underscores how aggressively federal agencies are now testing the limits of AI procurement, safety rules, and vendor trust.

What the court decided

The appeals court did not decide whether Anthropic’s AI is inherently unsafe. Instead, it agreed that the Defense Department could keep excluding the company from parts of its supply chain after Anthropic refused to accept a contract term the Pentagon considered essential.

That distinction mattered to the majority. The judges framed the dispute as a contract and procurement issue, not an attack on Anthropic’s broader political position on AI regulation. In their view, the government acted within its authority when it decided Claude should not remain embedded in its systems without the protections the department wanted.

The majority said the Pentagon had strong support for concluding that continued use of Claude in department systems created a national-security risk, and it rejected Anthropic’s claim that the government was punishing the company for its views rather than its contract position.

Anthropic has argued that it would not permit government use of its models for autonomous weapons or domestic surveillance. Defense Secretary Pete Hegseth has described that refusal as itself a serious security concern, widening the rift between the company and the Trump administration.

Why the Pentagon labeled Anthropic a risk

The Pentagon’s decision stemmed from separate supply-chain authorities that let the government limit vendors it believes create unacceptable exposure. In Anthropic’s case, the Defense Department sought to remove Claude from military and other federal systems by this month after concluding the company’s restrictions on use were incompatible with the department’s needs.

That position reflects a broader tension in AI contracting: the government wants access to high-performing models, but vendors increasingly want to control where and how their systems are deployed. Anthropic’s refusal to allow certain military and surveillance uses put it on a collision course with a defense establishment that values operational flexibility and vendor reliability.

The court said the department was not retaliating against Anthropic for advocating tighter AI regulation. Rather, it said the Pentagon excluded the company because it would not agree to a term the agency viewed as necessary.

How the case reached the appeals court

The dispute has moved through more than one courtroom because the Pentagon relied on separate legal pathways to impose the restrictions. One challenge was litigated in San Francisco, where a federal judge tossed out one of the labels in March and later reaffirmed that ruling. The other designation went to the D.C. Circuit, which on Friday let it stand.

The result is a split legal picture: one government action has been struck down, while another remains in force. Both sides are still likely to pursue appeals, meaning the final outcome could take years to settle.

How did Anthropic respond?

Anthropic said it still believes it is right and is weighing its next legal steps. According to company spokesperson Danielle Cohen, the firm remains confident in its position and is considering all options, including an appeal to the full D.C. Circuit or to the Supreme Court.

The company has not publicly detailed the full financial impact of the Pentagon’s actions, though it previously said the designation hurt revenue by making some customers wary of dealing with a business the government had effectively sidelined. That reputational drag may be particularly sensitive now, as Anthropic seeks to project momentum in the market.

The company has been talking up stronger sales recently and is widely seen as moving toward a potential initial public offering later this year. Any lingering government restrictions could complicate investor narratives about growth, trust, and regulatory exposure.

Why this matters for Anthropic’s business

The practical stakes go beyond one contract. Federal exclusion can shape how corporate buyers, partners, and investors perceive a model provider. A company that is locked out of parts of the U.S. government may find it harder to win procurement deals, or even to reassure enterprise customers that its product is broadly accepted.

• It can reduce direct government revenue opportunities.
• It may create reputational friction with commercial clients.
• It could affect IPO timing and valuation conversations.
• It may encourage competitors to pitch themselves as more government-friendly.

What does this mean for the Pentagon?

The ruling gives the Defense Department more room to keep steering away from Claude while it searches for alternatives. But the Pentagon has not provided clear updates on how far it has progressed in replacing Anthropic’s model or whether those alternatives are fully ready for comparable use.

Potential substitutes include models associated with SpaceX’s Grok, Google’s Gemini, and OpenAI’s GPT systems. The department has not said whether one model will replace Claude across the board or whether different tools will be used for different missions and workflows.

That uncertainty is important because government AI adoption is still in an experimental stage. Agencies want speed, accuracy, and flexibility, but they also want contractual control and policy alignment. When those aims clash, vendor selection becomes as much a legal issue as a technical one.

Why are Google and OpenAI part of the conversation?

Google and OpenAI matter because they are among the companies most likely to benefit if the Pentagon continues to move away from Anthropic. Their models are being discussed as possible alternatives, even as some employees inside those companies have objected to military deals on ethical grounds.

Those internal objections have not stopped the companies from pursuing government partnerships. Both firms have said serving the U.S. government is an important part of their work, presenting those contracts as a responsible way to support public-sector needs rather than an endorsement of every downstream use.

The contrast with Anthropic is striking: where Anthropic has drawn a harder line around military and domestic-surveillance applications, its larger rivals have been more willing to work with federal agencies. That difference may become a competitive variable if AI procurement continues to expand.

How the judges framed the free-speech argument

The appeals panel also rejected Anthropic’s claim that its constitutional rights were violated. The majority said the government complied with procedure and that the dispute was not about silencing Anthropic’s views on AI policy.

Instead, the judges said the Pentagon acted because Anthropic would not agree to a contract provision the department considered essential. In other words, the court viewed the case as a procurement disagreement, not a First Amendment crackdown.

That framing is significant because companies in the AI sector increasingly use safety commitments as both policy and branding. If a customer demands broader rights to use a model than the vendor wants to provide, the dispute can quickly become a legal argument about whether the government is punishing the company for its stance or simply refusing to buy on those terms.

What the dissent suggests

The final ruling came down 2-1, indicating at least one judge saw room for Anthropic’s arguments. While the panel’s earlier hearing suggested skepticism on both sides, the split result also shows that the legal line between procurement discretion and constitutional burden is not entirely settled.

That uncertainty may help explain why both Anthropic and the government are likely to keep fighting. Each side has an incentive to lock in a precedent before the broader legal landscape around AI procurement hardens.

Timeline of the Anthropic-Pentagon dispute

The legal fight has unfolded quickly, but its consequences may last much longer. Here is a simplified timeline of the major developments so far:

1. Early 2026: The Pentagon identifies Anthropic as a supply-chain risk under separate legal authorities.
2. Spring 2026: Anthropic challenges the action in court and argues the government overstepped.
3. March 2026: A federal judge in San Francisco strikes down one of the designations.
4. April 2026: The D.C. Circuit panel declines to temporarily block the remaining designation.
5. Friday: The appeals court upholds the Pentagon’s decision by a 2-1 vote.

Why this ruling could echo beyond Anthropic

The case is bigger than one company because it tests how far the U.S. government can go in conditioning access to federal contracts on model behavior and vendor commitments. If the Pentagon’s approach survives further appeals, other agencies may feel emboldened to demand similar concessions from AI suppliers.

That could affect the shape of AI governance in practice. Companies may be forced to choose between restrictive safety policies and lucrative public-sector contracts. Governments, meanwhile, may gain leverage to insist on broader usage rights, lower restrictions, or more direct assurances about model behavior.

For investors, the ruling is another reminder that AI companies face not only technical and competitive risk but also regulatory and procurement risk. For policymakers, it shows how quickly a dispute about model use can turn into a foundational question about national security and commercial autonomy.

What happens next?

Anthropic still has legal options, but the company now faces a tougher path. It could ask the full D.C. Circuit to rehear the case, take the matter to the Supreme Court, or continue litigating the separate San Francisco dispute. None of those routes offers a fast resolution.

In the meantime, the Pentagon can continue to avoid Claude, and the broader Trump administration can keep distancing itself from Anthropic’s tools. That may not be the final word, but it is a meaningful one for a company trying to maintain momentum while navigating an unusually public clash with the federal government.

For now, the practical outcome is simple: Anthropic remains on the outside of at least part of the defense marketplace, and the Pentagon retains the legal room to keep it there.