Framework
Bounded Agency in AI Systems— A Framework for AI Governance
What Is Bounded Agency?
Bounded agency is the condition in which an AI system acts with real autonomy, but only within limits that people set, can see, can change and can enforce in time to matter. When AI agency is bounded, human agency is preserved: people remain the authors of the decisions made on their behalf.
The Bounded Agency framework describes what that condition requires, in six primitives that must hold at once: authority boundaries, tempo matching, safe mode and rollback, intervention rights, reauthorization and accountability. Grounded in institutional economics and organizational theory, it explains how authority migrates into automated systems, and gives organizations a way to measure whether their AI remains bounded as it operates at machine speed.
What the BAL Score™ Measures
The Bounded Agency Limit (BAL) Score™ is a proprietary diagnostic metric that quantifies an organization's AI governance capacity across structural dimensions. Scored from 100 to 600, it identifies where governance architecture is intact, where it is drifting, and where intervention has become structurally infeasible.
The online BAL Score™ is an indicative, self-reported diagnostic: it shows where to look. Whether controls actually work is established separately, by testing them, and reported as Pass, Fail or Unverified for each of the six primitives. A strong score never offsets a failed control.
The Four Zones
Zones describe indicative, self-reported results. They are not findings about whether controls work, which only tested evidence can show.
Responses are consistent with intact governance capacity: the mechanisms, authority structures and intervention capability to govern AI systems are reported as in place.
Responses suggest autonomous authority is outpacing intervention capacity. Governance mechanisms exist but appear structurally insufficient to match the speed and scope of AI decision-making.
Responses indicate structural exposure that needs remediation, with governance gaps that create material risk under normal operating conditions.
Responses indicate a risk that AI systems are operating beyond reliable governance, with institutional intervention capacity structurally compromised.
Academic Foundation
The framework is set out in the book Bounded Agency and in a research paper available on SSRN. The evidence standard behind assessments is being published openly by the KnowGence Research Foundation.
Read the Paper on SSRN →