The US military relied on Anthropic's Claude AI for intelligence assessments and target identification during Saturday's strikes on Iran, even as President Trump had just ordered federal agencies to immediately cease using the tool — exposing a stark contradiction at the heart of Washington's AI policy.