Anthropic said that it can't "in good conscience" comply with a Pentagon edict to remove guardrails on its AI, despite a Pentagon ultimatum.