The Frontier Has No Inputs: "Pacing the Frontier" and the Development Layer of American AI Governance

Abstract: American AI governance defines the frontier without its inputs. In July 2026, over a thousand frontier AI workers petitioned Washington to lead an international effort pacing automated research. Their statement omitted the training data behind that capability. Read as a governance document, the petition demarcates who speaks and what is governed. The Great American AI Act discussion draft draws the same boundary, defining development without training data and routing disputes to federal copyright litigation. The public receives no seat. Framing society as a passive beneficiary, builders place development governance above state authority through claims of uncontainable competition. Meanwhile, the White House Accord on Super Intelligence abandons public oversight, leaving evaluation to internal developer controls. Deployment alone remains with the states. Between unexamined data extraction and the downstream injuries of deployment, the law carves out an insulated frontier. Keywords: AI Governance, Federal Preemption, Federalism, Development Layer, Great American AI Act, Pacing the Frontier, Training Data, Data Provenance, Consent, Copyright, Fair Use, Representative Claim, Epistemic Communities, Products Liability, Deployment, White House Accord on Super Intelligence, Political Economy of AI, State Laboratories, Algorithmic Extraction, Tort Preemption