Three-Dimensional Typology of Agency for Advanced AI Systems in Development

Researchers have developed multiple multidimensional frameworks between July 2025 and June 2026 for systematically categorizing agency in advanced AI systems. The typologies range from eight-dimensional approaches to three-part governance frameworks and aim to make autonomy, decision authority, and accountability in AI systems measurable.

Eight-Dimensional Typology for Agentic AI

A paper published on arXiv on July 7, 2025 developed a typology with eight dimensions for agentic AI systems. The framework defines cognitive and environment-related agency in an ordinal structure and enables standardized mapping of AI capabilities. According to the publication, the framework is intended to facilitate identification of differences between systems and support informed decision-making when selecting and integrating agentic AI solutions.

Three-A Framework for Dimensional Governance

A framework published on June 19, 2026 establishes dimensional governance as an approach that tracks three core parameters: Decision Authority, Process Autonomy, and Accountability. The model describes how these dimensions are dynamically distributed across human-AI relationships and adapt to changing contexts. The paper appeared as arXiv:2505.11579.

Philosophical Foundational Dimensions: Intentionality, Rationality, Explainability

A separate study published on March 5, 2026 identified three fundamental dimensions for phenotyping agency: Intentionality, Rationality, and Explainability. According to the paper, these dimensions enable both qualifying a system as an agent and explaining its actions. The publication appeared as arXiv:2603.04746.

Cognitive Architecture as an Agency Dimension

A taxonomy published on January 18, 2026 describes the cognitive architecture dimension and how agents conduct reasoning. The paper distinguishes between early systems with linear planning loops such as ReAct, hierarchical structures with tree-search methods for complex problems (such as Tree of Thoughts), and recursive decomposition approaches like ReAcTree. According to the publication, architectural diversity influences the ability of agentic AI systems to manage increasingly complex tasks (arXiv:2601.12560).

Human-AI Trust as a Governance Dimension

A research paper from March 5, 2026 identified six dimensions of Human-AI Trust: Evaluative Attitudes, Relational Interaction, Cognitive Learning, Explanatory Guidance, Collective Coordination, and Operational Control. According to the publication, these dimensions frame governance requirements for safe human-AI collaboration (arXiv:2603.04746).

Continuous Dimensions of Autonomy

Earlier research from 2013 conceptualized autonomous capabilities along continuous dimensions such as Self-directedness and Self-sufficiency. A paper that appeared in revised form on April 25, 2026 references these dimensions and describes how they adapt dynamically to changing contexts and enable more fluid conceptions of AI agency (arXiv:2505.01651).

Practical Application of Typologies

The developed typologies enable standardized comparisons between agentic AI systems based on defined dimensions. They support scenario-based selection of system characteristics for specific applications and dynamic adaptation of governance structures, such as through context-dependent distribution of authority and accountability.