No Safe Tier

This paper, the first of a sequence under the Aden Effect framework, examines a condition that the established literature on AI in education and AI literacy is structurally unequipped to address: the arrival of generative artificial intelligence as the first and only epistemic authority in environments that lack the verification infrastructure against which its output could be assessed. The dominant AI-literacy frameworks have converged on a sound principle, that literacy consists less in fluent prompting than in the user's judgment, responsibility, and retained agency. This paper argues that the principle, however correct, presupposes a foundation, a curriculum, a reachable trusted source, an institution standing behind the credential, that is present in the settings where the frameworks are produced and absent in the settings this paper examines, and that where the foundation is absent the prescription does not merely become harder to follow but ceases to apply, since the instruction to verify presupposes a held truth the most vulnerable user does not possess. The paper advances a single thesis in two registers. The symmetry register, drawing on contemporary hallucination measurement and on the analysis of fabrication as a trained rather than incidental behavior, establishes that confident fabrication is endemic across the entire price and capability spectrum, that the most capable reasoning models do not escape it and on grounded tasks may exhibit it more, and that every model encodes the epistemic hierarchies of its training corpus in ways the user cannot see; there is no safe tier and no neutral origin. The Matthew register, drawing on measured tokenization disparity and its coupling to reduced accuracy, and on evidence from South Africa and India, establishes that this endemic condition does its greatest harm precisely where verification cost is highest. Positioned within the literature on AI and epistemic injustice, the paper's contribution is to supply the empirical coupling that literature gestures toward but rarely measure, and to draw the structural conclusion that because the hazard cannot be escaped by purchasing a better tool, the human educator is not a transitional safeguard but a permanent locus of accountability. The paper discloses that it was drafted in cooperation with AI systems, and that those systems produced confident fabrications during its preparation which are treated as evidence.