"The Paradox of AI-Generated Research Without Human Editorial Review: A Case Study of Recurring Errors, Scholarly Reliability, and the Limitations of Artificial Intelligence Self-Auditing"

The Paradox of AI-Generated Research Without Human Editorial Review: A Case Study of Recurring Errors, Scholarly Reliability, and the Limitations of Artificial Intelligence Self-Auditing This independent observational research paper by Yeshayahu Ben-Yehudah presents the third stage of an ongoing investigation into artificial intelligence reliability, instruction fidelity, scholarly accuracy, and the limitations of AI-assisted self-evaluation. The study examines a deliberate research procedure in which an AI-generated scholarly manuscript was published without prior human editorial review. The published document was subsequently submitted to the same artificial intelligence system for examination, resulting in the identification of additional editorial deficiencies, bibliographic limitations, and methodological concerns. The investigation also documents repeated instruction deviations during the selection of research topics for Academia.edu. Despite explicit corrections, the AI repeatedly altered established keyword-entry formats, recommended terminology that the platform reportedly did not accept, and transferred additional verification responsibilities to the researcher. Building upon two previously published studies, the paper examines the distinction between AI-generated content, successful publication, independent scholarly verification, and the reliability of AI self-auditing. The research introduces proposed measures for postpublication defect identification, AI self-audit precision and recall, correction persistence, platform-specific terminology suitability, and human correction burden. The study further emphasizes human self-governance through the Ben-Yehudah Six-Pillar Process Architecture: STILLNESS, AWARENESS, OBSERVATION, DISCERNMENT, INTEGRATION, and TRANSMISSION. This publication forms part of a continuing observational research series in which successive papers receive distinct DOI identifiers to preserve the chronological development of documented incidents, corrections, and research findings. The investigation presents observational evidence and theoretical evaluation methods rather than controlled experimental results or confirmed internal software defects. Author: Yeshayahu Ben-Yehudah Publisher: THE BEN-YEHUDAH INSTITUTE Publication Date: October 30, 2026 Research Series: Investigating AI System Defects — Paper III DOI: 10.5281/zenodo.23827103 ORCID ID: 0009-0002-9256-1093 Permanent DOI Link: https://doi.org/10.5281/zenodo.23827103 Artificial Intelligence Disclosure: Artificial intelligence assisted with the organization, drafting, language refinement, source identification, and editorial development of this publication. The original research questions, observations, identification of AI-generated errors, deliberate publication procedure, and Six-Pillar Process Architecture are attributed to Yeshayahu Ben-Yehudah. The AI system assisting with the manuscript is also the subject of the investigation. No accompanying image was generated for this publication.