A review article titled "Cellular functions of spermatogonial stem cells in relation to JAK/STAT signaling pathway" was published online on February 13, 2024, in Frontiers in Cell and Developmental Biology (impact factor 5.5, CAS Zone 2 journal). Within three days, the paper was retracted due to concerns over illustrations generated by artificial intelligence.

The corresponding author, Hao M, is affiliated with the Red Cross Hospital of Xi'an (affiliated with Xi'an Jiaotong University School of Medicine) and serves as dean of its Spinal Disease Institute. Co-authors include Guo Y and Dong M from the same institution.

While the text reviewed the interplay between spermatogonial stem cells (SSCs) and the JAK/STAT pathway—highlighting SSC capabilities in self-renewal, multipotent differentiation, tissue regeneration, immunomodulation, and relevance to regenerative medicine—the figures diverged sharply from scientific norms.

Critics noted several implausible visuals:

• A rodent illustration depicted four large testes and an exaggerated penis, whereas rats normally have two testes.
• A signal-transduction graphic resembled a circuit board interspersed with pasta shapes and doughnuts.
• Cellular diagrams bore strong resemblance to toppings on a sausage pizza.

These images exhibited telltale artifacts of generative AI, prompting immediate scrutiny. The publisher, Frontiers, stated in a February 16 announceemnt that one reviewer had raised valid concerns about the figures and requested revisions, but the authors did not respond. An internal investigation is underway to determine why the failure to address reviewer feedback went unaddressed.

Illustrative examples of the problematic figures were shared by image forensics expert Elisabeth Bik, who remarked that the case reflects a troubling naïveté among journals, editors, and reviewers when handling AI-generated content. She warned that if such crude images passed peer review, more convincing AI graphics may already be present in scientific literature.

Technical implications for AI use in research papers

Chinese legislative discussions around academic integrity, including a draft degree law submitted in August 2023, propose revoking degrees for submitting AI-generated thesis content or granting unauthorized degrees. However, AI is not universally prohibited; permissible uses include data collection, literature search, outline drafting, and formatting assistance—tasks with low originality demands.

Key guidelines emerging from publishers:

1. Authorship policy — Papers where AI tools are listed as authors are rejected. Authors must be natural persons accountable for factual accuracy, completeness, and scientific rigor.
2. Disclosure requirement — If AI contributed to data analysis, figure creation, algorithm generation, text polishing, or similar tasks, the methods section must specify:
   • The AI tool(s) used
   • Detailed usage steps
   • Nature and extent of contribution
3. Academic misconduct — Failure to disclose AI involvement or submission of predominantly AI-generated content leads to rejection or retraction.

AI applications currently occupy a gray area in scholarly publishing. Transparent and disciplined use, accompanied by explicit documentation, is essential for maintaining research integrity.

Example of compliant disclosure in methods section

### Methods
Data categorization was assisted by GPTResearcher v1.2 to collate relevant publications from PubMed and Scopus. Image elements depicting pathway interactions were initially drafted using DALL·E 3, then manually verified for anatomical accuracy. Text refinement for clarity was performed with GrammarAI, focusing on results interpretation. All AI-generated components underwent validation to ensure alignment with experimental observations.

Example of noncompliant scenario

A manuscript submitted with detailed schematics of cellular structures suspiciously resembling AI art styles, without declaration of AI use in figure preparation, was flagged post-publication. Review revealed consistent visual patterns characteristic of diffusion models, leading to retraction under breach of disclosure policies.