Lanzhou University Probes Professor Over AI Watermark in Research Paper
Lanzhou University has launched an investigation into a professor from its College of Chemistry and Chemical Engineering after a research paper published in a top-tier scientific journal was found to contain a visible watermark from ByteDance’s Doubao AI tool, according to The Paper. The incident has ignited widespread debate in China’s academic community about the boundaries of artificial intelligence use in scholarly publishing.
The Discovery
The paper, titled “Electrostatic Gating Synergizing with Size Sieving: A MOF-Based Nanofiltration Membrane for Mitigating the Permeability–Selectivity Trade-Off in Lithium Recovery,” was published in Journal of Membrane Science (JMS), Volume 747, article 125315. JMS is an Elsevier-published journal ranked JCR Q1 and classified as a Chinese Academy of Sciences Zone 1 Top journal, with an impact factor of approximately 8.8.
A researcher in the scientific community noticed that Figure 2(d) in the published paper contained the text watermark “豆包AI生成” (Doubao AI Generated) in the corner of a chart, indicating the image had been processed using ByteDance’s AI tool without removing the platform’s native watermark. The discovery was shared on social media around June 24-25, quickly spreading across Chinese academic and media platforms.
University and Journal Response
On June 27, Lanzhou University issued an official statement confirming it had established a special investigation team. The university stated it maintains a “zero tolerance” policy toward research misconduct and will handle the matter seriously based on investigation findings.
That same day, Journal of Membrane Science published a statement via its WeChat official account, as reported by The Paper, acknowledging receipt of a complaint and confirming it had initiated an investigation following Elsevier’s publishing ethics policies. The complaint involves potential falsification, tampering, or improper processing of images in the paper, as well as whether AI usage disclosure complied with journal requirements.
Preprint Comparison Reveals Nuanced Picture
Analysis comparing the published version with a preprint posted on SSRN in December 2025 reveals a more nuanced situation than simple data fabrication. The preprint version of the same figure was clean and watermark-free. The only difference between the preprint and published versions was the replacement of sample labels from “PEI”/“PEI-UiO-67-NH₂” to “PA”/“PA-UiO-67-NH₂.” The FTIR experimental curves remained identical, suggesting core experimental data was not altered.
According to analysis by Sohu, the authors likely used Doubao AI for a simple text editing task — changing labels on a figure — and inadvertently left the watermark on the final image. This suggests the violation was more about procedural negligence than deliberate data fabrication.
AI Policy Violations
Elsevier has published specific guidelines on generative AI use in scientific images. AI tools may be used for flowcharts, conceptual diagrams, data visualization, and graphical abstracts, but they may not be used to generate or modify images representing original experimental data. Authors must disclose AI tool usage transparently.
Journal of Membrane Science’s author guidelines explicitly prohibit the use of generative AI or AI-assisted tools to create or modify submitted paper images. Even if the authors only used Doubao for simple text label modifications without altering core experimental data, this action violated the journal’s submission guidelines.
Peer Review System Under Scrutiny
The most significant implication of this case, as QQ News noted, is the exposure of weaknesses in the peer review process. The Doubao watermark passed through author self-review, co-author review, peer reviewer scrutiny, journal editor review, and the publisher’s final check — none of which detected the obvious watermark.
This raises serious questions about the thoroughness of peer review at even top-tier journals and has prompted calls for the adoption of AI-detection tools in the review process.
Broader Implications
The case represents a new frontier in China’s intensifying crackdown on academic misconduct: AI-related violations. In recent years, multiple Chinese universities have conducted retroactive checks on published papers, and several doctoral degrees have been revoked due to plagiarism and data fabrication.
This incident may prompt Chinese universities to develop clearer guidelines on AI use in research and accelerate the adoption of AI-detection tools in journal peer review processes. For researchers worldwide, it serves as a cautionary tale about the importance of proper disclosure and the risks of careless AI use in academic publishing.
What’s Next
Both Lanzhou University and Journal of Membrane Science have launched investigations. Key questions remain: What specific disciplinary action will the university take? Will the paper be retracted? And how will this case influence policies on AI use in research across Chinese academia? The answers will likely shape academic integrity standards in the AI era for years to come.