Sapiens in Control: A Six-Dimension Framework for Evaluating Control Failures in Advanced AI Agents

Abstract

Sapiens in Control, formerly Human in Control, proposes a six-dimension Human Control Model for reviewing human authority, containment, intervention, and oversight in advanced AI agents. Version 1.5 builds on v1.4 and adds source-qualified analysis of unintended actions against real external systems, a supplementary Philadelphia false-tip review, and separate discovery, containment, and notification intervals. It examines monitoring evaluations, independent-control proposals, implementation vulnerability disclosures, governance responses, and prospective risk assessments. An evidence-disposition appendix accounts for the daily safety review topics while preserving source-access limits and uncorroborated leads. The original five pilot cases, six dimensions, qualitative severity definitions, and prior corrections are retained. HIC-006 is supplementary and its severity remains unresolved. Proposed tests are not completed studies. This is an unreviewed conceptual working paper without completed independent human coding or empirical validation. AI tools assisted source retrieval, editorial comparison, drafting, and document preparation; this revision is not private-log verification, independent reproduction, or peer review. Previous releases remain preserved under the same concept DOI.