Court rules Anthropic safety refusal is Pentagon supply risk

This case represents a novel intersection of AI safety governance and federal procurement law, specifically regarding how 'supply-chain risk' is defined in the context of dual-use technology. Historically, supply-chain risk assessments focused on foreign ownership, cybersecurity vulnerabilities, or component provenance. The reclassification of voluntary safety guardrails as supply-chain defects creates a new category of procurement liability centered on functional availability rather than technical integrity. This mirrors earlier disputes in the biotechnology sector where contractors faced exclusion for refusing to perform certain classified research, though those cases typically involved classified information handling rather than public safety policies. The precedent also aligns with broader administrative law trends favoring agency deference in national security matters, limiting judicial review of executive branch procurement determinations. Previous challenges to military contractor exclusions have rarely succeeded when framed as readiness issues, suggesting this ruling extends established administrative law principles to the emerging domain of AI safety restrictions rather than creating entirely new legal doctrine.

On September 25, 2026, the U.S. Court of Appeals for the D.C. Circuit issued a 2-1 ruling upholding the Department of Defense’s designation of AI laboratory Anthropic as a supply-chain risk. The decision affirmed the Pentagon's authority to classify corporate safety policies—specifically refusals to enable lethal autonomous weapons or mass surveillance—as unacceptable vulnerabilities in national security procurement. According to reporting by Reuters and AP News, the dispute originated in February 2026 when President Donald Trump and Defense Secretary Pete Hegseth publicly accused Anthropic of endangering national security through its usage restrictions on the Claude model. Following these accusations, the DoD formally designated Anthropic a supply-chain risk, effectively blacklisting the company from federal defense contracts.

Anthropic appealed this designation, arguing that its prohibitions on lethal weaponry were core safety features rather than contractual defects or performance failures. The company contended that treating ethical guardrails as procurement liabilities violated its rights and misinterpreted supply-chain statutes. However, the divided appeals panel sided with the Trump administration. As reported by Wired, the majority found that the Pentagon’s concerns regarding battlefield readiness were reasonable under existing procurement law. The court determined that an AI vendor’s inability or unwillingness to support specific military applications constituted a functional gap in the defense supply chain, regardless of whether that limitation stemmed from technical incapacity or voluntary policy.

Julie Orb confirmed via Bluesky on September 28 that the federal court had upheld the classification, noting that 'the safeguard became the liability.' This phrasing encapsulates the central tension of the case: safety mechanisms traditionally viewed as responsible AI development were legally reclassified as operational deficiencies. The court document (Case 26-1049) references the social media denunciations by President Trump and Secretary Hegseth as part of the procedural history leading to the removal process, though the legal ruling itself rested on administrative deference to the DoD’s readiness assessments.

The Department of Defense has not publicly commented on whether this supply-chain risk interpretation will be applied retroactively or prospectively to other AI vendors with similar safety policies. The ruling establishes a precedent where corporate ethics guidelines can be adjudicated as national security liabilities. Anthropic maintained throughout the proceedings that prohibiting lethal weapon use was a non-negotiable safety standard, but the court’s decision suggests that in the context of federal defense contracting, such standards may now be subject to override by executive branch readiness determinations. The 2-1 split indicates significant judicial disagreement on the balance between corporate autonomy in safety design and executive authority over defense procurement criteria.

AI laboratories with safety-first business models risk exclusion from federal defense contracting worth billions annually. Companies must now weigh maintaining ethical restrictions against market access, potentially forcing policy modifications or market exit. The Department of Defense gains administrative leverage to override corporate safety determinations through procurement mechanisms rather than legislation. Defense contractors without restrictive safety policies may gain competitive advantage. Broader AI safety ecosystem faces uncertainty as voluntary guardrails become legally contestable in national security contexts. Magnitude depends on whether ruling spurs similar actions across federal agencies or remains isolated to current DoD interpretation.