The verification collapse cascade proceeds from rational nonverification through epistemic sedimentation to the dissolution of institutional reality
Year: 2026
Type: article
Abstract: As artificial intelligence outputs surpass human discriminative capacity, society confronts a novel threat to its capacity to determine truth from falsehood, a threat that requires no intentional attacker. Unlike accounts that locate truth’s erosion in deliberate distortion, the core threat identified here is that truth loses its institutional determinability even where no one is deceiving anyone. This paper proposes the Verification Collapse Cascade (VCC), a three-stage process through which, under the conditions specified here, society’s capacity to determine truth from falsehood degrades along a path that later correction does not retrace. Stage 1, Verification Evaporation, describes rational abandonment of verification when the return on checking AI outputs falls toward zero. Stage 2, Epistemic Sedimentation, describes the accumulation of unverified knowledge through erasure failure and existence fabrication, which the mechanisms that could have prevented it cannot afterwards undo. Stage 3, Institutional Reality Dissolution, describes how fact-determining institutions (scientific, judicial, journalistic, democratic) lose the internal capacity to render fact-fiction distinctions socially binding. The VCC is unified by a phase-transition metaphor reflecting causal dependency: no evaporation, no sedimentation; no sedimentation, no dissolution. Conventional countermeasures presuppose the will to verify and lose efficacy once evaporation advances. The paper proposes that the preservation of human verification capacity and the integrity of the shared knowledge base belong among the primary variables of AI governance, alongside output accuracy.
Source: Discover Artificial Intelligence
Author: Kenji Yamada
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $1,590
OpenAlex ID: W7155631033
