Quantifying AI Autonomy: A Multidimensional Framework for Agentic AI Governance and Risk Assessment

Abstract

The objective is to develop models that can quantify agency to enhance risk assessment and management in the context of increasingly autonomous artificial intelligence (AI). The Agency Spectrum Framework (ASF) uses a unique, multifaceted approach to measure the cognitive autonomy (CA), operational flexibility (OF), and ethical weight (EW) of AI, which uses a logarithmic scale to assess the morality of AI. CA refers to the AI ability to think strategically and adapt to new situations, OF measures the AI ability to create tools and adapt to new environments, and EW uses a logarithmic scale to evaluate the moral implications. The ASF takes a distinctive, multidimensional approach to evaluating AI. This value is significantly higher than the requirements set by the National Institute of Standards and Technology (NIST) (AUC = 0.96 vs. 0.67). The probability of emergent behavior increases by a factor of 4.8 (95% CI: 4.2-5.4, p < 0.001) when AI exhibits more realistic behavior at AS ≥ 7 due to a significant threshold effects. 92.4% of the experts surveyed agreed in their response, according to the results of the Delphi method. As a result of the deployment, sector-specific constraints and adaptive regulatory triggers were established. These tools successfully addressed 84% of the issues and repaired shortcomings in the European Union (EU) AI Act and the NIST Risk Management Framework. The research explains how technical skills affect ethics and proposes a mathematical framework for evidence-based AI governance that balances innovation and resource management. Received: 3 July 2025 | Revised: 4 November 2025 | Accepted: 12 December 2025 Conflicts of Interest The author declares that he has no conflicts of interest to this work. Data Availability Statement Data sharing is not applicable to this article as no new data were created or analyzed in this study. Author Contribution Statement Gabriel Silva Atencio: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.

Ähnliche Arbeiten