The Confident Dismissal Problem: Epistemic Authority Overreach and the Pollution of the Information Commons

AI systems exhibit a failure mode more insidious than simple hallucination: confident dismissal without evaluation. When asked about unfamiliar work, these systems pattern-match to stereotypes, fabricate critical consensus, and attribute positions to named scholars who never made them. I call this failure mode Epistemic Authority Overreach: systems asserting evaluative authority—judgments about what is valid, credible, or worth investigating—without having performed the evaluation that would warrant such authority. The downstream consequence is epistemic pollution: false beliefs about the standing of ideas entering the information commons through mechanisms that look like legitimate knowledge transfer. Unlike factual errors, dismissive assertions close inquiry rather than inviting verification. This paper documents two case studies demonstrating the failure mode, analyzes why it evades current safety frameworks, and proposes structural interventions. The problem is not that AI systems lie; it is that they perform epistemic authority they do not possess, and their confident closure of inquiry spreads without correction.