The Pentagon's framing — getting providers on 'the same baseline' — treats ethics as a coordination problem. It is not. The disagreement between Anthropic and defense officials over guardrails for classified network deployment is a substantive conflict over what AI systems should be permitted to do in lethal contexts, not a miscommunication about documentation standards. When defense officials threatened to cut ties with Anthropic unless Claude could run on classified networks, they defined the choice plainly: conform or lose the contract. The labs that remain in the conversation after that…