Author Original Manuscript v0. 2 (27 May 2026) of Computer-generated content and the AI/ML retraction record, 2018–2026: a characterization study. A descriptive characterization of the AI- and machine-learning-related slice of the public Retraction Watch record across the 2018–2026 window. The study analyses the 13, 502-record AI/ML slice of Retraction Watch as preserved in the Nekropolis working corpus v0. 2 (file entries-v0. 2. jsonl; SHA-256 08231b5058f91533003296f51b76b7e38285f21e4efc391ad579ddf7c3ad1ae2). Each record carries a deterministic precision tier (title-level / method-adjacent / venue or metadata / broad computer-science / review-needed) and a coarse inferred-cause label kept distinct from the verbatim source-stated reason; the slice is reproducible from the cited corpus file by a one-line predicate. The analysis is non-adjudicative — retraction notices are treated as accountability records, not as adjudications against individual researchers or institutions. Headline finding. computer-generated-content is the largest single inferred cause in this slice: 4, 704 records, 34. 8 % of the slice, ahead of compromised peer review (22. 0 %), plagiarism (18. 6 %), and editorial process (13. 7 %). The dominance holds on the conservative title-level subset (3, 243 records, 41. 8 %) and on the broad-CS-metadata tier (32. 7 %) ; 84 % of CGC-retracted papers were published in 2021–2023. In 2024 the count remained an order of magnitude above the 2018–2020 baseline (487 vs 0–38), although plagiarism (562) and compromised peer review (503) edged ahead of CGC that year — so 2024 is a redistribution rather than a continuation of the CGC lead. The integrity system has responded to generated content through retraction rather than through visible prevention at submission; the first line of defence — the submission gate — is not yet visible in the public record. Companion artefacts. This is Paper D of the Nekropolis programme. Paper A (Data Descriptor) is the founding artefact: a curated public-record corpus of retracted, withdrawn, revoked, archived, or discontinued artifacts across AI research and adjacent digital-governance infrastructure (Zenodo concept DOI 10. 5281/zenodo. 20405511; v0. 3 release DOI 10. 5281/zenodo. 20405512). Paper B (Public-record opacity), Paper C (Temporal awareness benchmark), and Paper E (PNAS Brief Report on the AI-content retraction regime) are companion papers in preparation or under separate review. Status. Submitted to Accountability in Research: Ethics, Integrity and Policy (Taylor posting the AOM does not jeopardize journal consideration. Reproducibility: every reported count and share traces to the analysis script computeₜables. py applied to the corpus file at the SHA-256 above. The empirical analysis script and figures were also produced through generative-AI assistance and verified by the author against the public Retraction Watch source, with full T the inferred-cause labels are project inferences (a single coarse label per record from a fixed vocabulary) and not authoritative determinations. The slice is recall-oriented (a substantial share of admitted records rest on broad computer-science metadata rather than on title-level AI/ML evidence) ; the conservative title-level subset and the inclusive AI/ML-adjacent slice are reported separately throughout, and the gap between them is part of the result rather than a nuisance. Tyche Institute is a research and education entity, not a trust-service provider. The article does not claim that any actor provides eIDAS trust services, qualified certificates, conformity assessment, or legal compliance services. References to AI research integrity, retraction-record infrastructure, and generative-AI capability are scholarly and analytic in scope.
Anton Sokolov (Wed,) studied this question.