Discovery as Delegated Authority: Governing AI-Mediated Meaning in Public Institutions

Discovery systems are treated as infrastructure: neutral machinery that locates material a reader then evaluates. AI-mediated discovery breaks that assumption. A system that infers which ideas are related, ranks arguments against one another, or summarises a body of work is not retrieving — it is interpreting, and it does so under the institution's name. This paper argues that AI-mediated discovery is an allocation of interpretive authority, and that public institutions are making that allocation implicitly. It identifies the resulting failure mode — authority exercised without accountability attached — and specifies a governance framework over four elements: decision rights, provenance and traceability, accountability boundaries, and legitimacy safeguards. The argument is addressed to boards and executives rather than to system designers. Its test is not whether a system performs well but whether an institution can answer four questions about it: why did it surface this, on what basis was it ranked, can it be reviewed and corrected, and who holds decision rights over its operation. Where those cannot be answered, the institution is already exercising interpretive authority it has not authorised. The paper closes by arguing for standards ahead of tooling, on the grounds that governance which arrives after deployment is retrospective by construction.