CADRE: Charter, Authority, Decision Rights, Records, Escalation. The Governance Function Specified as Five Artifacts

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

A board grants a mandate for AI governance and the institution announces a function. Some months later a supervisor asks a question the announcement cannot answer: under what authority was that system released, who exercised it, and where is the record. That gap is not a documentation gap. It is that the mandate was never converted into anything the institution can produce on demand. This specification defines CADRE, the AI Governance Operating Model: the structure, decision rights and cadence of the AI governance function, specified as five artifacts an auditor can test. Charter, Authority, Decision rights, Records, Escalation. Each term is drawn from the framework's registry entry rather than fitted to a letter, and the specification gives the derivation term by term. It is entry REG-07 of the Defensible AI Framework Registry, and that entry governs its relationships. The purpose is narrow and load-bearing: to convert a mandate into authority that is exercised on a schedule and recorded, which is the difference between a governance function and a governance intention. Two rules distinguish it from an organization chart, and both exist because the failure they prevent is the ordinary one. Exactly one person is accountable per decision, and an assignment naming a committee as accountable does not satisfy the requirement — a committee named as accountable is an accountability that dissolves the moment it is questioned. And a cadence with no meeting records is not a cadence: the record is the artifact, including the record of a meeting cancelled and why. The calendar is not evidence. CADRE holds a position no other framework in this family holds. It is the sole authority source in the register, granting authority to model risk management, data governance, vendor risk, incident response and the deployment spine, and the published register carries no edge whose target is CADRE. What returns to it is an escalation. The figure in this record draws that asymmetry, which is why the cascade has no return path. The specification states what it is not at the front rather than in an appendix, so a reader can establish the boundary before investing in the detail. It is not an organization chart or a recommendation about reporting lines; not a specification of competencies, training or credentials; not a replacement for the three lines model, which it positions the AI function within rather than restates; not a replacement for ISO/IEC 42001 or the NIST AI Risk Management Framework; not a certification scheme, and no accredited body assesses against it; and not a statement of regulatory compliance, which only a supervisor can make. Section 10 states what has not been tested. No institution unconnected to the author has been observed operating it, and no third party has published a result obtained under it: whether a chartered office with named decision rights surfaces a risk earlier than a competent committee is a fact about institutions, and no institution has supplied it. No trademark clearance search has been completed for the name, the specification asserts nothing about availability or registrability, and the symbol used throughout is the unregistered-mark symbol, which asserts a claim rather than a registration. No charter, assignment, cadence or escalation templates are published, so two institutions operating CADRE will produce artifacts that are individually conformant and not comparable. And the role set is calibrated to institutions large enough to staff it. The name is a rename, not a new framework. AI Governance Operating Model was retired as the name on 1 September 2026 and retained as the descriptor; the registry identifier is unchanged, because a rename records the former name and must not change the identifier. Work published under the former name remains citable, and a former name is not an error in the work that used it.