Anthropic pushed a two-sentence rule into Claude Code with no toggle, no changelog and no way to audit it. The rule is defensible - the delivery is not, and it is the fastest way to lose the AI race to a model nobody can switch off remotely.