The Pentagon should be using open models, not closed ones by OpenAI/Anthropic/xAI. The entire discussion of what Anthropic wants is therefore moot.