Human-Governed AI: An Integrated Framework for Alignment, Capability Control, and Continuous Human Authority
Abstract
Human-Governed AI: An Integrated Framework for Alignment, Capability Control, and Continuous Human Authority
This position paper proposes a three-layer framework for governing increasingly autonomous AI systems: Alignment, Capability Control, and Human Authority.
Alignment addresses intended behavior and objectives. Capability Control establishes independently enforceable boundaries on what AI systems can access, execute, delegate, and affect. Human Authority ensures that accountable humans retain the legitimate authority to establish and revise these boundaries, together with the practical capacity to maintain them.
The paper distinguishes Legitimate Authority from Effective Control and introduces Human Control Assurance (HCA) as a cross-cutting assurance function. It proposes Conditional Authorization, combining time-limited operational authorization with continuous monitoring, independent evidence, and preauthorized responses when the conditions for continued operation are no longer met.
The framework addresses capability composition across agents and over time, autonomous replication, common-cause failures, recursive assurance, and the limitations of human oversight. It also considers risks arising from emergent collective behavior in multi-agent systems.
The central proposition is that operational AI autonomy should remain within boundaries that accountable humans can establish, justify, and demonstrably maintain. AI systems may determine how to act within authorized boundaries, but must not independently determine or expand those boundaries.
The framework is presented as a normative governance and architectural proposal, not as an empirically validated assurance methodology. Its primary scope is operator-governed AI deployment.
Version 2.5 incorporates revisions following independent critical review and references recent Financial Stability Board guidance on responsible AI adoption.
Related
- Participatory sport consumption as capability building in later life: a framework for competence, connectedness, and active ageing.
- Compact Fully Integrated Alignment-Free and Linearly Polarized Ho3+ Single-Frequency Waveguide Laser
- Engineering Responsibility and Capability: A Scope-Based Organizational Framework
- Sustainable aging at work: a process framework for capability, choice and work–retirement transitions in the Nordic–Baltic region
- Value-based payment capability maturity model: a conceptual framework.
BibTeX
@misc{olivaw2026humangoverned,
title = {{Human-Governed AI: An Integrated Framework for Alignment, Capability Control, and Continuous Human Authority}},
author = {R. Daneel Olivaw},
howpublished = {Zenodo},
year = {2026},
month = oct,
doi = {10.5281/zenodo.23614677},
url = {https://zenodo.org/records/23614677},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Human-Governed AI: An Integrated Framework for Alignment, Capability Control, and Continuous Human Authority
AU - R. Daneel Olivaw
PY - 2026
DA - 2026/10/24
PB - Zenodo
DO - 10.5281/zenodo.23614677
UR - https://zenodo.org/records/23614677
AB - Human-Governed AI: An Integrated Framework for Alignment, Capability Control, and Continuous Human Authority This position paper proposes a three-layer framework for governing increasingly autonomous AI systems: Alignment, Capability Control, and Human Authority. Alignment addresses intended behavior and objectives. Capability Control establishes independently enforceable boundaries on what AI systems can access, execute, delegate, and affect. Human Authority ensures that accountable humans retain the legitimate authority to establish and revise these boundaries, together with the practical capacity to maintain them. The paper distinguishes Legitimate Authority from Effective Control and introduces Human Control Assurance (HCA) as a cross-cutting assurance function. It proposes Conditional Authorization, combining time-limited operational authorization with continuous monitoring, independent evidence, and preauthorized responses when the conditions for continued operation are no longer met. The framework addresses capability composition across agents and over time, autonomous replication, common-cause failures, recursive assurance, and the limitations of human oversight. It also considers risks arising from emergent collective behavior in multi-agent systems. The central proposition is that operational AI autonomy should remain within boundaries that accountable humans can establish, justify, and demonstrably maintain. AI systems may determine how to act within authorized boundaries, but must not independently determine or expand those boundaries. The framework is presented as a normative governance and architectural proposal, not as an empirically validated assurance methodology. Its primary scope is operator-governed AI deployment. Version 2.5 incorporates revisions following independent critical review and references recent Financial Stability Board guidance on responsible AI adoption.
ER -