Towards Adaptive Categories: Dimensional Governance for Agentic AI

Researchers propose a dimensional governance framework for AI systems that tracks decision authority, process autonomy, and accountability across human-AI relationships rather than relying on static risk categories. This adaptive approach enables proactive risk management by monitoring system movement toward critical thresholds, offering a more flexible alternative to traditional categorical governance as AI capabilities evolve.

The paper addresses a fundamental challenge in AI governance: traditional categorical frameworks assume stable, predictable systems with clear risk boundaries, but modern AI agents built on foundation models and multi-agent architectures defy these assumptions. As systems become more autonomous and capable, fixed classifications become brittle and reactive rather than protective.

The dimensional governance approach represents a meaningful shift in regulatory thinking. Rather than assigning an AI system to a risk tier or autonomy level, the framework monitors how authority, autonomy, and accountability distribute dynamically. This allows regulators and developers to detect when systems approach critical thresholds—say, when a model's decision-making autonomy increases while human oversight decreases—and trigger pre-emptive controls before risks materialize. The 3As model provides explicit, measurable dimensions that can be tracked across heterogeneous systems.

For developers and AI companies, this framework offers conceptual clarity: instead of meeting categorical checkboxes, teams understand specific governance dimensions they must balance. For regulators and policymakers, dimensional tracking enables context-responsive oversight that adapts to rapid capability gains without requiring constant framework rewrites.

The approach bridges a critical gap in AI policy discourse. Markets and innovation thrive with clear rules; static categories often stifle both through either over-restriction or dangerous loopholes. By making governance dimensions explicit and observable, this framework reduces uncertainty while preserving flexibility for legitimate innovation. However, implementation requires robust monitoring infrastructure and consensus on threshold definitions—challenges that remain largely unresolved.

• →Dimensional governance tracks decision authority, autonomy, and accountability instead of rigid risk categories
• →The framework enables proactive risk detection when systems approach critical governance thresholds
• →Adaptive categorization allows governance rules to evolve as AI capabilities advance
• →Static categorical frameworks increasingly fail for dynamic multi-agent AI systems
• →Implementation requires standardized monitoring infrastructure and threshold consensus across stakeholders