Human Control of Advanced AI Agents: A Six-Dimension Framework and Evidence Review

This standalone working paper proposes a six-dimension Human Control Model for reviewing meaningful human control of advanced AI agents. It examines authorization, containment, intervention, oversight, propagation, and situational grounding through five illustrative pilot cases and supplementary public-source analysis. The evidence review includes unintended actions against real external systems, monitoring studies, software disclosures, governance responses, and prospective assessments, with explicit source limits. Real-world effects, controlled evaluations, and simulations are distinguished. The supplementary Philadelphia false-tip case has unresolved severity. This edition reissues project research with presentation and metadata changes; it does not report new experiments. The framework remains conceptual, unreviewed, and empirically unvalidated. The author publishes under a pseudonym for personal privacy. AI assisted source retrieval, editorial comparison, drafting, and document preparation.