Court rules Pentagon can blacklist Anthropic for refusing to enable Claude features

The U.S. Court of Appeals for the D.C. Circuit upheld the Defense Department’s blacklisting of Anthropic for refusing certain military uses of Claude, even without malicious intent.

The majority said overly constrained models could fail mid-operation while unconstrained models could hallucinate lethal targets—and that the Secretary did not exceed authority under the Supply Chain Security framework at issue. Petitions for review were denied.

A Northern District of California judge had found a parallel designation unlawful under a narrower statute that requires adversary-style bad motive. Different laws, different courts, split outcomes.

This is the governance collision we keep saying is coming: a lab’s safety limits versus a state’s demand for unconstrained tools. The court framed both risks as sobering and then deferred to the Secretary’s balance.

The statutory split matters. Under 10 U.S.C. § 3252, the California court required something like sabotage or malicious subversion—and found Anthropic had no such motive. Under 41 U.S.C. § 4713, the D.C. Circuit read “any person” and “deny” broadly enough that refusing contractually demanded features counts as supply-chain risk. Same company, same stance, opposite lawful results.

We’re for labs that refuse autonomous weapons and domestic surveillance hooks. We’re also not naive: governments will use procurement and designation power to force the issue, and a 2-1 panel with two Trump-appointed judges is a political as well as legal signal.

If this stands, “safety policy” becomes a contract term you can be blacklisted for. Anthropic says it may seek en banc or Supreme Court review. Commerce’s Howard Lutnick has claimed the relationship is “in tune.” The courtroom says otherwise for now.

Jon Brodkin, Ars Technica, September 25, 2026. Judges Katsas and Rao formed the majority; Judge Henderson dissented. The same court had denied an emergency stay in April.