Anthropic loses Pentagon blacklist appeal over Claude safeguards

The dispute centers on two limits Anthropic placed on Claude: no fully autonomous weapons and no mass domestic surveillance. In a Feb. 26 statement, CEO Dario Amodei said the company supported military and intelligence applications, citing reliability and human oversight as reasons for the weapons restriction, and pointing to existing work in intelligence analysis, operational planning and cyber. Anthropic said in February that the Pentagon had demanded it relax both restrictions to keep the relationship.

Dissenting Judge Karen LeCraft Henderson challenged the majority's reading, arguing the statute targets deceptive interference and disputing that a contractor's openly enforced restrictions count as that conduct. The ruling covers only one of two Pentagon designations; a California judge set aside the other on Aug. 27, and that order blocks broader restrictions. Anthropic said it disagreed with the decision and was considering further review on Sept. 25.

The stakes appear to hinge on which safeguards a company is willing to keep even at the cost of defense work, and, for the suppliers building around Claude, on the reach of the exclusion. OpenAI's account of its own Pentagon agreement is one attempt to pair safeguards with defense access, though how its protections would hold during a dispute over military use remains uncertain.