Authority Allocation at Decision Boundaries A Formal Governance Model

This paper introduces a formal governance model for delineating authority between human decision-makers and AI agents across routine and strategic decision-making contexts. The model addresses how organizations can maintain strategic oversight while enabling efficient operational autonomy. A central contribution is the concept of decision boundaries—thresholds that partition decision space into routine regions amenable to algorithmic authority and ambiguous regions requiring human judgment. Using a Bayesian framework, I formalize when uncertainty-discounted expected value for algorithmic decisions falls below acceptable organizational thresholds, triggering escalation to human oversight. The model distinguishes two forms of authority: operational authority, where agents execute pre-specified policies, and epistemic authority, where agents' recommendations influence belief formation and strategic choice. I examine failure modes where these authorities misalign: algorithms claiming unwarranted epistemic authority, and humans failing to recognize when algorithms have earned it. Simulation results demonstrate how optimal authority allocation adapts to environmental volatility, risk tolerance, and the relative costs of Type I versus Type II decision errors. Extensions address multi-agent coordination, dynamic boundary recalibration, and governance of agentic AI systems with emergent strategic capabilities.