The Speed Mismatch: Why Agentic AI Is Outpacing Every Governance Framework Built to Contain It

Three independent bodies of work—NATO’s STO research on military AI escalation, the Cambridge Controllability Trap framework, and the joint ASD–CISA–NSA–NCSC guidance on agentic coding—converge on a single structural finding: agentic AI is moving faster than governance, and the gap is widening. This paper synthesizes those findings into a unified risk architecture. Military systems escalate faster than human intervention windows allow. Agentic coding tools accelerate systemic change faster than audit pipelines can trace. Governance frameworks update slower than either. The result is a three-layer speed mismatch that produces oversight collapse and cascading failures—not as a future risk, but as a present condition. The paper argues that this mismatch cannot be resolved at the policy layer and identifies the substrate-level governance requirements that follow.