Nuance · 2025–26 Anthropic, the Pentagon & the Guardrails

The popular shorthand — "Anthropic handed a military deal to OpenAI" — is not quite the record. Anthropic held a $200M Pentagon contract (Claude was the first frontier model in classified networks). When the Pentagon demanded "any lawful use" language — removing Anthropic's bans on using Claude for mass domestic surveillance and fully autonomous weapons — Anthropic refused. The administration then blacklisted it as a "supply chain risk" (a label previously reserved for foreign adversaries) and OpenAI took the classified-systems deal, with its own CEO calling the timing "opportunistic and sloppy." A federal judge later blocked the designation. Read the guardrail fight however you like — the refusal is the documented part.