AI Generated Scientific Paper Passes IEEE Access Review, Raising Questions About Scholarly Integrity

A researcher prompted a large language model to draft a complete scientific manuscript, including hypotheses, code snippets, and prose, without any real experiments. The resulting document was submitted to IEEE Access, a multidisciplinary, open-access journal. The managing editor, Jenny Mahoney, reviewed the submission as part of a controlled test. The paper passed the journal’s editorial screening and was sent to peer reviewers. The experiment demonstrated that current editorial and peer-review procedures can be fooled by a well-crafted AI output.

The author employed a “vibe-research” approach, feeding the model natural-language prompts that guided it to generate plausible hypotheses, statistical analyses, and code. The process required minimal human intervention beyond selecting prompts and reviewing the output. The final manuscript mimicked the structure and tone typical of IEEE publications, including citations, figure captions, and methodological sections.

IEEE Access follows a standard editorial workflow: initial screening for scope and formatting, followed by assignment to subject editors and external reviewers. The journal relies on human judgment to detect plagiarism, data fabrication, and methodological flaws. No automated LLM-detection tools were employed during the test, allowing the AI-generated text to slip through.

The experiment underscores a growing risk: large language models can produce technically coherent, publication-ready manuscripts that may evade traditional review safeguards. If unchecked, such content could proliferate, diluting the quality of the scientific record and eroding trust in peer review. Publishers may need to integrate LLM-detection algorithms, enforce stricter disclosure policies, or adopt automated plagiarism checks that flag synthetic language patterns.

IEEE Access has announced plans to review its editorial policies and explore tooling that can detect AI-generated text. Other journals may follow suit, potentially adopting open-source LLM-detectors or collaborating with AI researchers to benchmark detection accuracy. The broader community faces a choice: either enhance existing workflows or develop new standards for AI-authored content.

Reports estimate that by 2025, 57% of academic articles show detectable LLM influence, up from 12% in 2023. In Nature Communications, over 12% of reviews were found to be entirely AI-generated in 2025. These figures suggest that AI-slop is already a significant fraction of scholarly output, and the IEEE Access case illustrates a concrete pathway for such content to enter the literature.

• Integrate LLM-detection APIs into manuscript submission portals.
• Require authors to disclose AI assistance in the methods section.
• Train reviewers to spot synthetic language patterns, such as repetitive phrasing or improbable data consistency.
• Collaborate with AI ethics groups to establish guidelines for acceptable AI use in research writing.

Publishers may need to allocate budget for new software tools and reviewer training. Funding agencies could mandate AI-disclosure clauses in grant proposals, adding administrative overhead. The cost of implementing robust detection systems may be offset by preserving the integrity of the scientific record.

• How effective are current LLM-detection tools against newer, more sophisticated models?
• Will journals enforce mandatory AI-disclosure policies, and how will compliance be verified?
• Can automated systems reliably distinguish between legitimate AI assistance and deceptive fabrication?