Governance Framework for Agentic AI Autonomy Levels
A new study on arXiv (2607.23438) introduces a governance framework that separates Allowed Autonomy Levels (AAL) from Autonomous Capability Levels (ACL) in AI systems. AAL is influenced by elements like risk management, oversight, and accountability, while ACL focuses on the technical abilities of the system. The autonomy levels range from reactive execution and decision support to supervised actions, goal-directed autonomy, and delegated operational authority. This research proposes a decision-making approach that considers risk when assessing allowed autonomy and explores how risk and accountability evolve across different levels of autonomy.
Key facts
- Paper on arXiv: 2607.23438
- Introduces governance framework for agentic AI
- Separates Allowed Autonomy Levels (AAL) from Autonomous Capability Levels (ACL)
- AAL defined by risk, oversight, and accountability
- ACL characterizes technical abilities
- Autonomy levels: reactive execution, decision support, supervised action, goal-directed autonomy, delegated operational authority
- Proposes risk-aware decision process for assigning allowed autonomy
- Analyzes evolution of risk and accountability across autonomy levels
Entities
Institutions
- arXiv