The American Society for Biochemistry and Molecular Biology has adopted explicit guidelines for AI-assisted writing in its journals, and authors who fail to follow them risk publishing fabricated citations or distorted technical terms that can damage their credibility. With papers increasingly containing AI-generated errors - including an estimated one in 277 biomedical papers with fabricated references - researchers need to know exactly where the boundaries are.
ASBMB rules: what AI can and cannot do
The ASBMB's editorial policies are clear on authorship: AI tools cannot be listed as authors. The policies require authors to be "accountable for all aspects of the work," and an AI tool cannot accept responsibility for inaccuracies because it simply generates output based on its programming. Human authors remain responsible for the accuracy of their manuscripts.
What AI can do is polish existing text. The policies allow authors to use AI to improve grammar or style, which can help researchers writing in a second language communicate more clearly. But AI may polish text only - it may not write any part of a manuscript.
Authors must disclose AI use in a separate section titled "Declaration of generative AI and AI-assisted technologies in the writing process." The declaration must name the tool used and state that the authors have reviewed the output and take responsibility for its content.
When AI produces figures
AI tools generally may not be used to create, manipulate, or refine manuscript images. The exception is when AI-generated images are produced as part of the experimental design itself - for example, a study using deep learning to improve medical imaging. In that case, authors must describe how the images were generated in sufficient detail for others to reproduce the work, just as they would for any experimental method.
This distinction matters for researchers using AI in fields like radiology or microscopy where machine learning is part of the science, not just the presentation.
Hallucinated citations and tortured phrases
AI language models generate text by predicting the most likely next word based on patterns in training data. When the model lacks the information needed to respond accurately, it fills in the gaps. This is called "hallucination," and it's a feature of AI that is unlikely to disappear.
Hallucinations are especially concerning in reference lists. Fabricated citations often look convincing enough to escape peer review, even including inaccurate or nonexistent DOIs. The source content notes that in biomedical literature, about one in 277 papers now contains fabricated references - sometimes called "aren'ticles." Because references may function as research data, fabricating them could constitute research misconduct.
AI-assisted writing can also produce "tortured phrases," where familiar scientific terms get distorted. These often arise when AI-generated text is run through plagiarism-avoidance software that replaces words with synonyms. Technical terms often generate absurd results like "bosom peril" instead of "breast cancer" or "polymerase chain response" instead of "polymerase chain reaction."
Verifying everything AI produces
Researchers who choose to work with AI tools must verify every AI-generated citation and phrase against the original source, especially in a reference list. The stakes extend beyond a single paper: unverified AI output can undermine an author's credibility and damage the scientific record.
For those building practical AI skills within research workflows, training programs like the AI for Research Scientists learning path offer guidance on using these tools without crossing integrity boundaries. More broadly, the AI for Science and Research resources cover managing AI in complex projects.
Why this matters for research scientists
Your publication record depends on trust. A single fabricated reference or distorted technical term can trigger retractions, funding reviews, or reputation damage that follow you through your career. Verify every AI-generated reference against the original source, keep AI tools within the grammatical improvement limit your journal permits, and disclose their use even when your journal may not require it.
Your membership also unlocks:
