Boundary Blindness Under Artificial Intelligence: Early Cross-Industry Findings on the Missing Decision-Evidence Layer

Abstract

Artificial intelligence does not create the gap this research traces across regulated markets. It converts that gap from a chronic condition the market tolerated for decades into an acute one it cannot tolerate. When a decision crosses a boundary between two organizations, two software systems, or two regulatory regimes, the data may cross, but the evidence of how the decision was reached does not. Who reviewed what, under what authority, against which rules, with what conclusion, and whether that conclusion still holds, stays behind. The receiving party gets the outcome, not the basis for relying on it. This paper calls the failure to see this single cause behind many separate symptoms boundary blindness. Drawing on seventy-one semi-structured interviews with senior practitioners and on independently published sources across seven sectors and five jurisdictions, the paper catalogues nine operational symptoms grouped into cross-firm, within-firm, and artificial-intelligence clusters. It documents a convergence cluster of sources that independently name the same architectural failure without referencing one another. It argues that the AI-governance regimes now in force largely certify the wrong thing, assessing systems rather than the decisions those systems produce, and that obligations run upward to supervisors or inward to a firm's own records rather than sideways to the counterparty actually relying on a determination. The paper proposes one architectural intervention: a portable, tamper-evident decision record, termed an acceptance signal, expressed as an openly licensed specification governed as a public good and freely implementable by any party. It sets out the empirical phase designed to test and, if warranted, falsify the diagnosis. This paper consolidates the artificial-intelligence dimension of a twelve-month research programme (October 2025 to October 2026) and generalises it beyond the insurance market in which the programme began. It presents findings, not a product, and makes no commercial claim. Companion papers and the research programme: https://www.veridafoundation.org/research