From Answers to Auditable Intelligence: A Governance-First Framework for Agentic AI in Strategic Research and Public Policy
Description
Agentic AI can draft a persuasive answer in seconds. Strategic research, however, is not short of answers — it is short of answers whose evidence can be inspected, whose uncertainty is visible, and whose conclusions a named human being is willing to own. This paper designs the workflow that produces them. It proposes a six-layer governance framework for agentic AI in strategic research and public policy (question framing; evidence acquisition; claim-level traceability; adversarial review; human decision gates; post-hoc monitoring) and an evaluation protocol built on measurable dimensions including sourced-claim rate, citation precision and recall, evidence coverage, fabrication rate and calibration. Status: this is a working paper, not peer reviewed. Its framework has not been validated by a blinded pilot; the paper cites a companion published citation-audit study and reports a first internal implementation verification run.