Anthropic Seeks Court Rehearing in Pentagon AI Safety Dispute

Anthropic has asked a US federal appeals court to reconsider a ruling that upheld the Pentagon’s decision to designate the artificial intelligence company a national-security supply-chain risk. The petition, filed on October 13, 2026, challenges a September decision by a divided panel of the US Court of Appeals for the District of Columbia Circuit. The dispute centres on Anthropic’s refusal to remove certain safety restrictions from its Claude AI models, particularly restrictions against their use in fully autonomous weapons and mass domestic surveillance.

The case has become an important legal test of how the US government can regulate access to AI technology used in defence operations. The Pentagon argues that a technology supplier’s restrictions could interfere with military requirements, while Anthropic contends that the designation was unlawful, lacked adequate procedural safeguards and was retaliation for its public position on AI safety. The appeals court’s September decision favoured the government, but Anthropic’s request for a rehearing seeks another review of the legal questions involved.

Key Takeaways

• Anthropic seeks a rehearing: The company wants the US Court of Appeals for the DC Circuit to reconsider its September ruling.
• The earlier decision was 2–1: A divided appeals-court panel upheld the Pentagon’s supply-chain risk designation.
• AI safety is at the centre of the dispute: Anthropic has resisted allowing Claude to be used for fully autonomous weapons and mass domestic surveillance.
• The Pentagon cites military reliability: The court majority accepted the government’s argument that model restrictions could interfere with military operations.
• A separate California case reached a different result: A federal judge there struck down a parallel designation under a different legal authority.
• The outcome could affect AI contractors: The case may influence how defence agencies and AI companies negotiate acceptable uses, restrictions and oversight.

Why Anthropic Is Seeking a Rehearing

Anthropic’s petition challenges the legal reasoning behind the September appeals-court decision. The company argues that the majority interpreted a federal supply-chain security law too broadly, applying a mechanism associated with protecting government systems against supply-chain threats to an American AI company because of restrictions built into its models.

According to Anthropic’s position, the Pentagon acted without an adequate pre-designation process and despite evidence that the decision was retaliatory. The company argues that its safety policies should not automatically make it a national-security risk simply because those policies limit certain military applications.

The petition asks the court to revisit the decision. A rehearing is not the same as a new trial or an automatic reversal. The court must decide whether to reconsider the earlier ruling, and filing the petition does not itself remove the designation or restore the company’s access to defence-related work.

The legal dispute is especially consequential because it concerns both the government’s authority over its suppliers and the limits of a company’s ability to set conditions on how its technology is used.

What Is the Pentagon’s Security-Risk Designation?

A supply-chain risk designation allows the US government to take action against technology or suppliers considered a threat to government information systems and national security. In this case, the Pentagon invoked the Federal Acquisition Supply Chain Security Act framework to exclude Anthropic’s Claude AI models from its supply chain.

The law is intended to help protect government systems from supply-chain threats. The dispute concerns how that authority applies when the supplier is a US AI company whose models contain restrictions on certain uses.

The Pentagon’s position is that Anthropic’s restrictions could prevent Claude from carrying out tasks the military considers authorised and necessary. The appeals-court majority agreed that the government had sufficient grounds under the applicable law to treat the restrictions as a potential supply-chain risk.

Anthropic, by contrast, argues that its safeguards are intended to prevent dangerous or inappropriate uses rather than expose military systems to hostile interference. The company also disputes the process and reasoning behind the government’s action.

The designation has implications beyond a conventional contract disagreement. It can affect whether defence contractors may use Anthropic’s products in work for the Pentagon, creating potential consequences for the company’s government business and relationships with other suppliers.

The Dispute Over Autonomous Weapons and Mass Surveillance

The conflict escalated after negotiations over the conditions for military use of Claude failed. Anthropic has maintained restrictions against fully autonomous weapons and mass domestic surveillance, two areas in which it believes current AI capabilities and safeguards raise serious concerns.

Autonomous weapons

Fully autonomous weapons are systems that can select and engage targets without a human making each individual targeting decision. The debate involves questions of reliability, accountability, human control and the consequences of errors in high-stakes environments.

Anthropic has argued that frontier AI systems are not sufficiently reliable for unrestricted use in such applications. The company has sought to retain limits on the uses it considers unsafe.

The Pentagon’s position is different. It argues that military users must be able to rely on technology to perform authorised tasks without a private supplier’s restrictions unexpectedly preventing essential operations.

The disagreement is therefore not simply about whether AI should be used by the military. It is also about who determines the permitted uses of a model, how those limits are enforced and whether a supplier can change or restrict functionality after a government agency has integrated the technology.

Mass domestic surveillance

Anthropic has also opposed the use of its models for mass domestic surveillance. The issue raises questions about privacy, civil liberties and how AI could be used to analyse large volumes of information about individuals.

AI can help process information at scale, but that capability also creates concerns about monitoring people without adequate legal safeguards. Anthropic’s restrictions reflect its position that certain uses should remain outside the scope of permitted applications.

The government and the company have disagreed over the implications of these restrictions for military and national-security operations. The court dispute will help determine how the relevant law applies to the Pentagon’s decision, rather than resolving every broader question about AI ethics or surveillance policy.

What the September Appeals-Court Ruling Said

In September 2026, a three-judge panel of the US Court of Appeals for the DC Circuit ruled 2–1 in favour of upholding the Pentagon’s designation.

The majority concluded that the government had sufficient legal grounds to determine that the restrictions embedded in Claude could pose a risk to military operations. It rejected Anthropic’s arguments that the designation was unauthorised, arbitrary and unconstitutional.

The majority’s reasoning focused on the potential consequences of restrictions on technology deployed in defence systems. It said the government could consider whether those restrictions might prevent a model from performing tasks that military officials regarded as contractually authorised and necessary.

The dissenting judge disagreed with the majority’s approach. The divided ruling left Anthropic with the option of seeking further judicial review, which it has now pursued through its rehearing petition.

The decision did not establish that Anthropic’s safety concerns were inherently illegitimate. Rather, it upheld the Pentagon’s authority under the legal framework considered by the court. The dispute over the company’s motives, the scope of the law and the proper process remains central to Anthropic’s challenge.

Why the Separate California Case Matters

The litigation has produced a different result in a separate federal case in California. In August 2026, a federal judge there struck down a parallel designation under a different legal authority, finding that the government’s action was unlawful.

The two cases concern related government measures against Anthropic, but they involve distinct legal provisions and proceedings. Their different outcomes should not be treated as a direct contradiction in which one court has simply overturned the other.

The DC Circuit ruling concerns the Pentagon’s use of the supply-chain security framework. The California case addressed a parallel designation under another law and the legal basis for the government’s action in that proceeding.

This distinction matters because the outcome of one case does not automatically determine the outcome of the other. The legal authorities, records and questions before the courts differ, even though both disputes arise from the same broader disagreement over AI safety restrictions and military use.

Anthropic can therefore continue to challenge the DC Circuit ruling while the implications of the California decision remain relevant to the wider dispute.

The Financial and Commercial Impact on Anthropic

The designation has potential consequences for Anthropic’s government business and its position in the defence technology market. The company has said the Pentagon’s actions have cost it billions of dollars in lost business.

The effects can extend beyond contracts directly held by Anthropic. Defence contractors may need to review whether they can use Claude in work connected to the Pentagon, while organisations supplying technology to the military may reconsider how they manage dependencies on AI providers.

The designation can also influence future negotiations. AI companies may seek clearer contractual language about permitted uses, safety restrictions, monitoring and the circumstances under which a government customer can demand changes to a model’s behaviour.

For Anthropic, the legal outcome could affect its access to defence-related opportunities and its ability to maintain its position on restricted uses. However, the financial consequences should be distinguished from confirmed accounting losses: the company’s statements about lost business do not mean that every potential contract has been cancelled or that a specific amount has already been recorded as a loss.

The dispute may also influence investor assessments of regulatory risk and the cost of serving government customers.

What the Case Means for the Wider AI Industry

The case raises questions for AI developers that supply models to governments, defence contractors and other high-stakes customers.

Companies building advanced AI systems increasingly establish acceptable-use policies and technical safeguards intended to limit dangerous applications. Government agencies, meanwhile, may require technology that meets operational needs and can be used under their own rules and procedures.

Those interests can conflict. If a model provider refuses to permit certain applications, a government customer may decide that the product is unsuitable for its needs. If the government uses legal authorities to exclude the provider, the company may argue that the decision exceeds statutory limits or punishes its policy position.

The outcome could affect how future contracts address these issues. AI suppliers may seek more precise terms covering safety controls, responsibility for deployment decisions, access to models and the handling of disagreements. Government customers may seek stronger assurances that systems will remain available for authorised operations.

The case could also encourage greater scrutiny of how governments distinguish between a genuine supply-chain threat and a disagreement over a supplier’s acceptable-use policy. The courts’ treatment of that distinction will be important for companies working at the intersection of AI, cybersecurity and national security.

What Happens Next?

The immediate next step is for the appeals court to consider Anthropic’s request for a rehearing. The filing does not guarantee that the court will agree to reconsider its decision, and it does not automatically suspend the effects of the existing ruling.

If the court grants a rehearing, it could revisit the legal reasoning and the arguments raised by the parties. If it declines, Anthropic may consider other available routes for review, subject to the applicable rules and deadlines.

The company will also have to navigate the practical consequences of the existing designation while the legal process continues. Defence contractors and government agencies will need to follow the rules that apply to their work, while Anthropic will continue to assess the implications for its contracts and business relationships.

The broader policy debate will continue regardless of the immediate procedural outcome. The case highlights the difficulty of aligning private AI safety policies with government requirements for military technology, especially when the systems involved are capable of supporting high-stakes operations.

The Bigger Picture

Anthropic’s dispute with the Pentagon is about more than one company’s access to defence contracts. It raises fundamental questions about how the government can use procurement and supply-chain laws to manage risks associated with advanced AI, and how much control private developers retain over the uses of their technology.

The case also demonstrates that AI safety policies are becoming part of commercial and legal negotiations. Developers may view restrictions as essential safeguards, while government customers may regard them as limitations on operational control. The legal framework governing those conflicts will influence future contracts, supplier relationships and the adoption of AI in national-security settings.

Looking Ahead

Anthropic’s request for a rehearing creates another stage in its legal challenge to the Pentagon’s security-risk designation. The appeals court’s September decision remains significant unless it is changed through further proceedings, and the rehearing petition itself does not restore the company’s position in defence-related work. The next key development will be whether the court agrees to reconsider the ruling.

Over the longer term, the dispute could shape how AI companies negotiate safeguards with governments and how federal agencies apply supply-chain security laws to domestic technology providers. The outcome may influence the balance between military operational requirements, supplier control over AI systems and safeguards against potentially harmful uses. Until the court takes further action, both the legal and commercial consequences remain uncertain.

Frequently Asked Questions

1. Why is Anthropic asking for a rehearing?

Anthropic argues that the appeals-court majority misinterpreted the supply-chain security law, that the Pentagon acted without adequate process and that the designation was retaliatory.

2. Why did the Pentagon designate Anthropic a security risk?

The Pentagon argued that restrictions embedded in Claude could interfere with military operations by preventing the model from performing tasks the department considered authorised and necessary.

3. What AI uses are at the centre of the dispute?

The disagreement centres on Anthropic’s restrictions on fully autonomous weapons and mass domestic surveillance.

4. Did the appeals court rule in Anthropic’s favour?

No. In September 2026, a DC Circuit panel ruled 2–1 to uphold the Pentagon’s designation. Anthropic is now seeking a rehearing.

5. Did another court reach a different conclusion?

Yes. A federal judge in California struck down a parallel designation under a different legal authority in August 2026. That separate ruling does not automatically overturn the DC Circuit decision.

6. Does seeking a rehearing immediately reverse the designation?

No. Filing a rehearing petition does not itself reverse the ruling or restore Anthropic’s access to Pentagon-related work. The court must decide whether to grant the request.