Irreversibility Constrained Decision Theory for Human and AI Systems: Moral Reversibility, Epistemic Constraints, and Harm Bounded Authority Allocation

Artificial intelligence (AI) systems are increasingly embedded in high stakes decision processes involving life, liberty, and large scale environmental risk. Existing governance approaches emphasize auditability, documentation, and procedural oversight but do not impose a principled constraint on AI authority in irreversible decision domains. This paper proposes a Core Ethical Boundary: AI systems must not hold final authority over decisions that can cause irreversible harm, including death, permanent disability, indefinite loss of liberty, or non recoverable ecological damage. This boundary is grounded in the concept of moral reversibility capacity, defined as the ability to bear responsibility, be sanctioned, and participate in moral repair capacities that are absent in artificial systems. Having established this constraint, the paper argues that human decision making must also be structurally bounded. Traditional appeals to experience based judgment are insufficient in rapidly evolving organizational environments where historical expertise loses predictive value. Unstructured human judgment introduces inconsistency, hidden subjectivity, and non auditable reasoning. The objective is therefore not to privilege either AI or human agents, but to allocate authority, accountability, and performance evaluation according to their functional strengths.