DC Circuit Upholds Pentagon's Anthropic Ban; Extends to Contractors on Defense Work
The ruling inverts the traditional supply-chain-risk standard. Normally, DoD excludes vendors for foreign ownership, espionage risk, or technical backdoors—concrete threats.
Here, the court accepted that Anthropic's safety training itself is a national-security risk because it constrains what the model will do. That sets a template: any AI company whose training includes refusals (which all major vendors now have) can be designated a supply-chain risk if those refusals conflict with a military use case. The second move: contractors must now choose between compliance (dropping Claude) and contract risk.
Pentagon CTO Emil Michael just won the ability to exclude a major AI vendor from the entire defense industrial base, not just government systems—a precedent that invites weaponization of "supply chain risk" designations against any AI company whose safety guardrails conflict with military use cases.
The mechanism is stark: Anthropic encodes restrictions into Claude that prevent certain tasks; the court found this itself grounds for national-security exclusion, regardless of whether those restrictions actually harm defense capability. Watch the Northern District of California ruling on appeal—if that court reverses and the DC panel stands, Anthropic faces a circuit split that forces Supreme Court review, but in the interim, every defense contractor must audit their AI dependencies and begin Claude migration or face contract compliance exposure.
Does the Pentagon's ban extend to contractors' use of Claude in non-classified systems (e.g., administrative, HR, unclassified research), or only classified and defense-adjacent work? The article says "defense contracts" but not the scope of systems covered.
Fault Lines
Strategic intelligence, synthesized daily — with a public track record. Every call graded against what actually happened.