Court upholds Pentagon’s ability to blacklist Anthropic over Claude feature dispute

The Department of Defense said it could blacklist the AI startup Anthropic after the company refused to turn on certain Claude features for a Pentagon project. A federal appeals court in Washington, D.C., agreed that the Pentagon’s action is allowed under a different supply‑chain law. The ruling came on September 25, 2026, and it leaves the blacklist in place while the lower court’s earlier decision is limited to a narrower statute.

How we got here

In August 2026 a judge in the Northern District of California found that the Pentagon’s blacklist violated 10 U.S.C. § 3252. That law limits “supply‑chain risk” to cases where an adversary deliberately sabotages or inserts malicious code. The judge said Anthropic had not shown any bad intent, so the blacklist under that rule was illegal.

The Pentagon, however, also cited 41 U.S.C. § 4713, a broader authority that does not require proof of malicious intent. Congress gave the D.C. Circuit exclusive jurisdiction to interpret that law for defense procurement. The appeals court said the lower court’s focus on intent was irrelevant under § 4713, and therefore the blacklist could stand.

What happens next

The Pentagon can continue to treat Anthropic as a “supply‑chain risk” under the broader statute. Anthropic may still try to challenge the decision, but any further appeal must go through the D.C. Circuit, not the district court. The case highlights how the government can use different legal definitions to restrict AI vendors it deems risky.

For now, Anthropic’s Claude model will remain unavailable for any Pentagon systems that rely on the blacklisted features. Other AI companies watching the case may adjust their contracts with the Defense Department to avoid similar designations.