This paper introduces the concept of a Pre-Critical Recursive Cutoff (PCR-C), a staged infrastructure control framework designed to reduce irreversibility risk in recursively self-improving or highly autonomous AI systems. Rather than focusing on output alignment or post-hoc safety constraints, PCR-C shifts the safety boundary to the infrastructural layer prior to critical recursive escalation. The framework defines a pre-critical region in which intervention, refusal authority, and external constraint mechanisms remain institutionally and technically viable. Beyond a certain threshold of capability coupling, external connectivity, and autonomous modification capacity, system trajectories may enter an irreversibility zone where meaningful human intervention becomes structurally ineffective. PCR-C proposes a layered cutoff mechanism based on measurable indicators related to recursive modification cycles, external actuation capability, and infrastructural integration. The objective is not to halt innovation but to introduce a staged control boundary that activates before loss-of-control dynamics become dominant. By reframing AI safety as an infrastructural governance problem rather than a purely behavioral alignment problem, this approach contributes a structural model for pre-emptive risk mitigation in advanced AI deployment contexts.