D.C. Circuit Court Ruling Redefines AI Safety as National Security Risk
For years, AI labs have promoted their safety measures as a competitive advantage, assuring the public that their technologies would not be misused. However, on September 25, 2026, the D.C. Circuit Court of Appeals issued a ruling that fundamentally challenges this notion. In the case of Anthropic PBC v. Department of War, the court upheld the Pentagon's classification of Anthropic as a supply chain risk under the Federal Acquisition Supply Chain Security Act (FASCSSA) § 4713. This ruling sets a precedent that a lab's voluntary safety measures can be legally considered a national security liability, irrespective of the company's intentions.
The ruling raises the question of how safety features can be perceived as security threats. The Pentagon's determination, made by Defense Secretary Pete Hegseth on March 3, 2026, stemmed from Anthropic's refusal to eliminate contractual restrictions on its AI model, Claude, which barred its use in lethal autonomous warfare and mass surveillance. The Secretary insisted on an "all lawful uses" clause, which Anthropic declined. The court's majority, comprising Judges Gregory G. Katsas and Neomi Rao, interpreted FASCSSA as encompassing risks associated with a product's operation, thereby granting the executive branch the authority to override a company's safety policies if they hinder military utility.
This ruling effectively allows the government to frame safety as a supply chain risk, compelling compliance without needing to demonstrate malicious intent. The court dismissed Anthropic's claims of First Amendment retaliation and due process violations, indicating that when AI development intersects with defense procurement, governmental operational requirements take precedence over a lab's ethical considerations.
The dissenting opinion from Judge Karen L. Henderson highlighted the structural implications of this ruling. She argued that FASCSSA should necessitate a prior designation of "significant risk" before a supply chain finding is made. Her concern was not merely technical; she pointed out that the majority's interpretation grants the Secretary extensive, unchecked power over a contractor's policy decisions. When the government can label any safety measure as a supply chain risk, the line between voluntary ethics and enforced compliance blurs.
Legal Conflicts Impacting AI Safety Standards
Additionally, this ruling creates a split in legal interpretations. A related case in the Northern District of California, overseen by Judge Rita Lin, previously deemed a similar designation of Anthropic unlawful. As a result, AI labs now face conflicting legal signals based on jurisdiction, which could encourage strategic forum shopping and undermine consistency in legal standards.
The timing of this ruling is critical, occurring just two days before Anthropic's deadline for provable inference, a key milestone linked to the company's public safety commitments. The ruling complicates how Anthropic can present its safety measures as a competitive advantage to investors, given that a federal appeals court has now classified that same safety as a liability in the context of defense supply chains. The challenge lies in reconciling the narrative of safety as an asset while the government's legal stance treats it as a vulnerability.
Filed & Cleared
This development unfolds amid a tumultuous period for the AI sector. The establishment of the SAFA safety standards body by Anthropic, Google, and OpenAI occurred just three days after OpenAI's second training halt. Furthermore, an antitrust lawsuit filed on September 18 in California alleges that coordinated safety slowdowns among these companies constitute an output-restricting cartel. Collectively, these events illustrate a landscape where AI labs' control over their safety narratives is rapidly diminishing, influenced by courts, regulators, and market dynamics.
Anthropic is currently considering further legal actions, including a potential en banc rehearing or a petition to the Supreme Court. However, the structural implications of this ruling are already evident: when safety measures conflict with state requirements, they can be legally categorized as defects. Consequently, the lab bears the reputational risk of its technology, while the government dictates operational parameters, leaving the public—who were promised safety—watching as the framework they relied upon is redefined as a supply chain vulnerability.
