The Empty Occupant: Answerability, Testimony, and the Limits of Machine-Mediated Authority

Debates about AI and responsibility have mostly asked whether automation opens responsibility gaps. This monograph argues that a more pressing failure lies elsewhere. An institution can remain fully responsible and accountable for its machine-mediated decisions, yet lose its claim to be relied on, because it issues more than it can answer for. The book develops a substrate-neutral account of Answerability as four public relations: attribution continuity, challenge-responsive reconsideration, repair competence, and standing-consequence. Three results follow. First, missing bearership is not defective ownership. An AI interface can occupy a role emptily while the institution behind it still owns the role's commitments. So there can be no global responsibility gap and still a genuine local Empty Occupancy at the interface. Second, machine output can carry testimonial force without constituting testimony. An institution may adopt that output as its own only when five conditions of Normative Sponsorship are met: epistemic access, authority, review bandwidth, independent rejection, and repair power. Third, because practical authority is defeasible, reliance must remain revisable while a claim operates. Obligations that mature without timely service accumulate as Deontic Debt. Where no credible plan could service issuance at a given scale, the institution is insolvent in a precise sense. It loses presumptive authority, while its duties of ownership and repair survive. Public-record analysis of Australia's Robodebt scheme yields a determinate Deontic Debt that case-by-case actor–forum accountability does not settle; the insolvency finding remains provisional. The Air Canada and DPD records leave interface standing indeterminate, and a stipulated appeals-agent case supplies a determinate Empty Occupancy verdict. TennCare, Post Office Horizon, and the Dutch childcare-benefits affair discipline the comparisons. The account is tested against Bovens, List and Pettit, the responsibility-gap sceptics, and Lazar, and it closes with the conditions under which it would be wrong.