Dark profabulation squared describes the condition in which an AI system's weaponised temporal projection to manufactured future deployed needs as present-day military operation. These become training signal for both its own systems and any opponent with a similar AI system. This creates the horror of a recursive AI system, creating a closed loop that no human authorised and no human can interrupt. Drawing on the coined concept of profabulation (Matthews, 2026), this paper traces its weaponised inversion through NATO climate-security doctrine, Palantir's AIP/Gotham pipeline, the Tomorrow.io satellite integration, and the FedRAMP IL4–IL6 authorisation hierarchy. A section on Algorithmic Mutual Assured Destruction (MAD) develops the mechanics by which autonomous targeting on one side structurally compels autonomous targeting on the other not as policy choice but as temporal mathematical performance. Empirical grounding draws on Payne's (2026) frontier model wargaming study, which demonstrated that AI systems never voluntarily de-escalate and that a model appearing safely restrained in open-ended scenarios may invert to near-Strategic Nuclear War under deadline pressure. The paper concludes with a methodological critique of the precision-recall metrics used to legitimise these systems, exposing the epistemic fraud at the heart of the dark profabulation loop. The counter-move; constitutive illegibility as survival resistance is reserved for a companion paper.
The full text of this paper is available as a PDF — use the button above to read it.