Federal Court Affirms Pentagon Ban on Anthropic's Claude in Defense Contracts
The D.C. Circuit Court of Appeals upheld the Pentagon's decision to exclude Anthropic's Claude from defense supply-chain work, ruling 2-1 that AI safety guardrails can be classified as supply-chain risk factors. Anthropic refused Pentagon contract terms that would have required removing restrictions on autonomous weapons and mass surveillance capabilities. The ruling applies specifically to Department of Defense systems and contractor work, not to ordinary commercial or consumer use of the technology.
The D.C. Circuit's decision represents a significant expansion of how federal procurement law interprets security risks. The court's majority determined that safety features built into AI systems—specifically Anthropic's refusal to disable safeguards against autonomous weapons and mass surveillance—constitute legitimate grounds for excluding a vendor from defense supply chains. This classification treats the presence of protective restrictions as a potential vulnerability rather than a strength, creating a novel legal precedent for how AI capabilities are evaluated in government contracting.
Anthropic's position centered on maintaining ethical boundaries that the company viewed as non-negotiable. The company declined Pentagon terms that would have required fundamental modifications to Claude's architecture, particularly restrictions preventing unrestricted autonomous weapons use and domestic surveillance applications. The split decision, with one judge dissenting, suggests ongoing judicial disagreement about where the line should be drawn between legitimate security concerns and restrictions rooted in ethical design philosophy.
The ruling may reshape how federal agencies evaluate AI vendors and contract negotiations with technology companies. Organizations selling to the Pentagon could face pressure to modify safety features to remain competitive, potentially influencing how AI systems are designed across the industry. Conversely, companies may increasingly view certain safeguards as non-negotiable business positions rather than contractual bargaining chips. The decision's long-term impact depends partly on whether other agencies adopt similar frameworks and whether competing jurisdictions develop alternative regulatory approaches to AI procurement.