The Decision No One Authored: The Answerability Gap in Generative AI

Can a person retain full control of an AI-shaped decision and still fail to author it? Consider a hypothetical pension fund whose generative system guides its allocations through a decade of good returns. Officers review proposals, can override anything, and sign every trade: every outward mark of control, oversight, and authorship is present. Yet in this case, the conception of risk under which the money moves remains a vendor's default: no one at the fund ever decides what risk should mean for these beneficiaries. The decision appears authored, though the fund never made the judgment it turns on. The natural question is whether responsibility has shifted to the machine or remains secured by the humans around it. Debates over consciousness, patiency, and machine agency address the first possibility; meaningful human control and oversight address the second. Neither settles whether anyone performed that prior judgment. I call this the answerability gap. It is deepest where the system is best: reliability promotes the unowned frame into more consequential decisions and dissolves the checking that would expose it. The paper gives an account of the missing act and of the adoption that makes an inherited frame answerably one's own.