Artificial intelligence has become increasingly embedded in the practices through which human beings acquire, organize, evaluate, communicate, and act upon information. The philosophical problem raised by this development is therefore not exhausted by the familiar observation that artificial intelligence sometimes produces false information. More fundamental is the question of whether systems that generate apparently knowledgeable discourse can participate appropriately in epistemic practices when they do not possess the kinds of beliefs, experiences, intentions, understanding, or testimonial commitments ordinarily associated with human knowers. This paper develops the concept of epistemic impropriety as a normative framework for examining that problem. Epistemic impropriety is distinguished from epistemic error and epistemic harm: an error concerns falsity or inadequate justification; a harm concerns damage suffered in one's capacity as a knower; impropriety concerns the violation or distortion of norms governing responsible inquiry, testimony, justification, interpretation, authority, and epistemic agency. The paper argues that the central epistemic danger of generative artificial intelligence lies in the possibility that linguistic plausibility may acquire the social authority normally associated with justified knowledge. This danger manifests itself in hallucination, the manufacture of apparent understanding, the illusion of comprehensiveness, the substitution of plausibility for justification, algorithmic opacity, automation bias, synthetic testimony, epistemic injustice, hermeneutical exclusion, epistemic infantilization, and the gradual displacement of human intellectual agency. At the same time, the paper rejects the simplistic conclusion that artificial intelligence is epistemically worthless. AI may function as an epistemic technology, a cognitive aid, and a participant in distributed practices of inquiry without thereby becoming an autonomous knower or epistemic authority in the full human sense. The appropriate philosophical response is consequently neither technological rejection nor uncritical enthusiasm, but the cultivation of a critical human–AI epistemic relationship governed by provenance, transparency, evidential calibration, contestability, pluralism, human judgment, and responsibility. The ultimate question is not merely whether artificial intelligence can generate answers, but whether its participation enables human beings to become better knowers.