Human in the Log: Public Evidence Chains for Public-Sector AI Oversight

Public authorities increasingly describe AI and automated systems as advisory, supportive, or subject to human oversight. Those assurances are difficult to evaluate when the record does not show how the system entered the workflow, what output it produced, what the human saw, and what review or repair was available. This article develops the concept of the “human in the log”: a record-centered account of oversight in public-sector AI and adjacent automated decision systems. The anchor case is Colombia’s Constitutional Court decision T-323/24, which rejected a due-process challenge on sequence-sensitive grounds: the judge’s legal reasoning and decision preceded the ChatGPT consultation. From that case, the article builds a public-evidence synthesis of legal, regulatory, audit, procurement, register, standards, and observatory records. The aim is not to rank countries or count incidents. It is to identify evidence functions that make oversight reconstructable: sequence, notice, transparency layer, human capacity, public visibility, retention, and repair. The article proposes a seven-part oversight record — entry point, input/output, human view, human act, reasons, contest path, and retention/repair — as a practical architecture for making human control inspectable without treating logging as a substitute for legality, fairness, or accountability. Supplementary materials include the global case matrix (Supplement A), the replication codebook (Supplement B), and the 315-row case ledger (Supplement C).