The Genetic Fallacy at Scale
The rapid normalization of generative AI produced two distinct epistemic problems that keep getting treated as one. The first is a problem of quality: language models can generate false, derivative, or merely polished material at enormous scale. The second is a problem of judgment: the causal fact of where a text came from is increasingly substituted for the work of evaluating what it actually claims. The paper's central move is naming the practice of rejecting a claim because it carries an AI-generated or AI-assisted label, without engaging its content, as the genetic fallacy running at institutional scale, and it works through AI-text detectors' documented vulnerability to false positives, distribution shift, and bias to show why automation dressed up as human oversight does not fix the problem. It closes with six proposed norms for legitimate epistemic governance, provenance modesty, content-responsive refutation, proportionality, contestability, auditability, and human answerability, the last of which fixes responsibility for a consequential epistemic decision on the person and institution that relied on the tool, never on the tool itself.