Randomized trial explores governance-aware AI architecture in enterprise systems, suggesting practical strategies for accountability.
Key Points
This study aims to develop a governance-aware architecture for agentic AI in enterprise engineering, addressing the need for accountability and auditable decision-making.
Developed a governance-aware reference architecture for agentic AI using design-science methodology.
Synthesized existing AI risk management standards and human-centered AI design literature to formulate a quantitative risk-control model.
Evaluated the architecture through illustrative scenarios in customer service, finance operations, HR screening, and supply planning.
Introduced the Governance-Aware Agentic AI Control Architecture with six integrated layers.
Defined three constructs: Productivity-Adjusted Residual Risk, Governance Debt, and Human Override Threshold to facilitate deployment decisions.
Demonstrated the model's ability to translate governance principles into measurable engineering checks.