Abstract
Generative AI has intensified scholarly publishing's need to distinguish formal accountability for a manuscript from the epistemic capacity to answer for it. Existing authorship guidance assigns responsibility to human authors regardless of AI involvement, but assigning responsibility does not establish that it can be meaningfully exercised. This paper addresses that gap by distinguishing nominal from substantive authorship responsibility and introducing the Duty to Defend (DtD): the obligation, internal to the status of scholarly authorship, to retain sufficient epistemic command of one's work to provide a reasonable account of its central claims, methods, evidence, and interpretations when legitimately challenged. Although the expectation that authors can defend their reasoning is already emerging in the literature, the normative question of why authorship itself carries such an obligation remains open. The paper derives this obligation from the reciprocal structure of scholarly credit rather than treating it as an additional compliance requirement and distinguishes it from an AI-detection standard: the relevant question is never whether AI was used, but whether the named author remains capable of standing behind the contribution attributed to them. A conceptual model then identifies three configurations — integrated, attenuated, and displaced authorship — describing how technological assistance can preserve or weaken the connection between credit and human intellectual agency. The paper translates this framework into a graded, trigger-based approach to editorial assurance, in which scrutiny escalates only as substantive scholarly concerns remain unresolved, and examines its own limits, including evidentiary difficulty, unequal capacity to demonstrate understanding, and the risk of converting a safeguard for authorship into a mechanism of suspicion.