The Trump Administration wants Anthropic’s A.I. model, Claude, to act like an obedient soldier; the tech firm argues that that could lead down a dangerous path.