Anthropic Refuses Pentagon Demand To Remove Safeguards

In January–February 2026, Anthropic, whose Claude model was deployed inside Pentagon classified networks via a Palantir partnership, refused a February 27 demand to remove safety guardrails allowing domestic surveillance and fully autonomous weapons. The refusal led to a federal blacklist and halted agency use, accelerated alternative contracts, and intensified concerns about oversight, escalation bias, and AI reliability in military decisions.