Human in Control: A Practical Framework for Evaluating Human Control of Advanced AI Agents
Year: 2026
Type: article
Abstract: AI incident databases and risk taxonomies are increasingly valuable, but they often answer broad questions about harm, misuse, or system failure rather than the narrower operational question of whether humans remain meaningfully in control of increasingly agentic systems. This paper proposes the Human Control Model, a six-dimension framework for analyzing agent-control failures: Intervention Durability, Technical Containment, Informed Delegation, Oversight Integrity, Control Contagion, and Situational Grounding. The framework separates behavioral severity from environment reality so that controlled evaluations are not presented as real-world incidents. This v1.1 revision updates the paper title to better distinguish the work while preserving the Human in Control project name, the existing Human Control Model, the six-dimension framework, the original version lineage, and the publication record family. The framework is conceptual and case-based, without inter-rater reliability testing or validated quantitative thresholds. The prepared local export updates the recoverable v1.0.1 manuscript title and metadata; the prior longer sandbox-only v1.1 manuscript was unavailable. Creator: Human in Control Project. License: CC BY 4.0. The deposit includes the PDF working paper, editable DOCX source, prepared revision notes, and metadata JSON.
Author: Sapiens in Control Project
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7215117397
FWCI:
Cites:
Cited by:
Related to: