Anthropic insists it won’t let the military use its systems for domestic mass surveillance or autonomous weapons, but the Pentagon doesn’t do the former and the latter is far beyond current capabilities. Anthropic also refuses to accede to the Pentagon’s demand for access to the AI system’s full capabilities. But any technically competent actor with access to an AI model can bypass its behavioral constraints. All models have been publicly jail-broken within days of launch. The confrontation is over a lock that any adversary can pick.