An assistant can be honest, obedient and caring, yet become the only route through which a person finds evidence, frames alternatives and prepares disagreement. The right to correct survives; the capacity to form a correction narrows. Responding to Jakub Pachocki's call for AI that retains honesty, integrity and love for humanity across unfamiliar situations and optimization pressure, this essay proposes a prior constraint: keep human judgment causally effective. I call this capacity causal authorship: human beings must remain able to form reasons and make them change what a system does, what it is for and whether it continues. This is not another value to add, but the constraint under which love, honesty and every other value must operate. An aligned disposition must resist reward without resisting correction. This is not a training algorithm or a complete solution to alignment. It distinguishes assistance from benevolent replacement and suggests tests for whether that disposition survives pressure without appropriating the human processes that could revise it.