The Integrity Frontier: Data Corruption, the Attribution Problem, and Verification as Deterrence by Denial Beneath the Nuclear Threshold

Nuclear deterrence has not been rendered obsolete by artificial intelligence. What has emerged instead is a layer of strategic competition that operates beneath the nuclear threshold, where the contested asset is the integrity of data and the characteristic instrument is corrosion rather than destruction. This paper names that layer the Integrity Frontier and consolidates its argument as the Integrity Frontier Doctrine. The paper argues that classical, punishment based deterrence weakens on this layer for a specific and well documented reason: it depends on attribution, and attacks that degrade data and learning environments resist attribution by design. Drawing on Nye's taxonomy of cyber deterrence, the paper shows that the viable response is not a stronger threat of retaliation but deterrence by denial, achieved through verifiable integrity. The argument is grounded in three constructs developed by the author, namely the Synthetic Data Contamination Index (SDCI), Integrity Debt, and a verification first architecture, and is anchored empirically in the peer reviewed finding that models trained recursively on synthetic data undergo measurable collapse (Shumailov et al., 2024). A documented civilian case, Moffatt v. Air Canada, illustrates the failure signature: a structurally sound system producing a harmful outcome because of corrupted inputs rather than flawed logic. The doctrine is then translated into institutional terms, aligned with the NIST AI Risk Management Framework and the data governance provisions of the European Union's Artificial Intelligence Act, and expressed as three implementable policy measures. The paper deliberately reasons from documented cases and peer reviewed evidence rather than from unverifiable claims about classified or operational military programs.

Перевод пока недоступен