Citation:

Giglio, Edoardo and Politi, Alessandro, Beyond the Statistical Centre: Allied Artificial Intelligence, Strategic Recognition, and the Two Failures of the Mean  (September 14, 2026). Available at SSRN: https://ssrn.com/abstract=7636359 or http://dx.doi.org/10.2139/ssrn.7636359

Abstract

This paper argues that the artificial-intelligence systems now being procured at scale across the Atlantic Alliance are configured for the analytic task at which they are structurally weakest: the recognition of the consequential outliers — novel attacks, abandoned hypotheses, cross-domain recombinations — that have historically decided cognitive-security and innovation outcomes. It advances three claims and concedes the limits of each. First, that strategic recognition comprises two distinct failures of the mean — a distributional failure (rare-event detection) and an abductive failure (cross-domain inference) — which share a consequence, strategic surprise, but not a mechanism. Second, that the dominant deployment topology, the sequential analytic pipeline, compoundsthe distributional failure multiplicatively in the depth of the chain, so that fielded systems are more tail-blind than the single-model evidence shows. Third, that the corrective — an inverse-intelligence capability — is therefore not a model to be invented but a verification pipeline to be deployed, and that its near-total absence from Alliance systems is the central capability gap. The paper grounds these claims in a peer-reviewed evidence base, a risk heatmap across seven architecture classes, a verified map of the failure classes onto fielded NATO and ministry-of-defence systems, the historical case of the Ufimtsev–Lockheed stealth asymmetry, and a formal treatment of daisychaining, and it submits the reformed thesis to a structured adversarial review. A closing section extends the diagnosis to a second, normative axis — the freezing of value-alignment into unratified 'constitutions' and the resulting gap in normative interoperability between allied systems — showing it to be the same failure of the mean transposed into the ethical register.

Keywords: artificial intelligence, strategic recognition, NATO, defence AI, model collapse, RLHF, distribution shift, AI governance, arms control, inverse intelligence

Citation:

Giglio, Edoardo and Politi, Alessandro, Beyond the Statistical Centre: Allied Artificial Intelligence, Strategic Recognition, and the Two Failures of the Mean  (September 14, 2026). Available at SSRN: https://ssrn.com/abstract=7636359 or http://dx.doi.org/10.2139/ssrn.7636359