Fault-Based Publication Ethics: The Case for Source Custody in an Era of AI Citation Contamination

Abstract

Fabricated citations in the biomedical literature increased approximately tenfold from 2023 to early 2026, reaching one in 277 papers (Topaz et al., 2026), and 98.4% of flagged papers remained in the literature without correction at the time of the audit. Manual enforcement has failed. The response is automation: citation-verification pipelines, AIdetection products, and platform-level sanctions triggered by algorithmic flags are now entering scholarly production workflows. Automated enforcement solves the volume problem but creates a new one. AI-detection tools carry documented false-positive rates, and those false positives fall disproportionately on non-native English speakers, earlycareer researchers, and scholars at under-resourced institutions. An author wrongly flagged for citation fabrication under the current framework has no standardized evidentiary record to produce in rebuttal, because no publication ethics system asks authors to document what they verified at the time of use or provides a mechanism for producing that documentation when a citation is challenged. This paper argues that publication systems need a fault-based integrity model that makes verification effort visible, and it makes five contributions toward that model: (a) it defines four postpublication failure modes (retraction, link rot, content drift, and exposed fabrication) that can invalidate a source after good-faith use; (b) it documents the laundering mechanism through which invalidated sources compound in downstream work, drawing on evidence that over 94% of post-retraction citations in biomedicine never mention the retraction (Hsiao & Schneider, 2021; Bakker et al., 2024); (c) it establishes that reference rot predates AI and affects 13% to 75% of web citations depending on field and age (Klein et al., 2014; Zittrain et al., 2014), positioning AI as an accelerant on an already unstable citation substrate; (d) it proposes a five-level fault ladder separating fabrication, failure to verify, negligent verification, downstream contamination, and source decay, with graduated consequences calibrated to each level; and (e) it proposes a Source Provenance Ledger, a private evidentiary record maintained by the author and producible on challenge, building on Glynn's (2025) full-text reference deposit proposal and following established precedents for voluntary documentation in law, medicine, journalism, and financial services. Adoption is distributed across three channels: author prerogative (the author decides which outputs warrant documentation), publisher requirement (journals may require that submitters maintain a ledger producible on request), and insurance condition (errors-and-omissions policies may condition coverage on source custody records). The paper does not excuse careless AI use and does not claim the ledger is validated. It presents a fault-based framework, an evidence base, a staged validation roadmap, and an open invitation for replication. All referenced governance frameworks are published opensource under Creative Commons at github.com/basilpuglisi/HAIA.

Keywords: Fault-Based Publication Ethics, Source Custody, Citation Contamination, Hallucinated Citations, Fabricated Citations, Reference Rot, Due Diligence Evidentiary Record, Publication Integrity, Retraction, Source Provenance Ledger