Nature Biotechnology's AI Policy: Springer Nature Rules Meet Biotech's AI-Heavy Workflow

Here's where Nature Biotechnology diverges in practice from, say, Nature Geoscience or Nature Astronomy. A substantial fraction of papers published in Nature Biotechnology use AI/ML as part of the research itself. Your paper might describe:

• A deep learning model for predicting protein-protein interactions
• A generative model for designing novel enzyme variants
• A reinforcement learning approach to optimizing bioreactor conditions
• An NLP pipeline for mining biomedical literature

None of this is covered by the AI manuscript preparation policy. The AI you used as a research tool belongs in your standard Methods section as methodology, not as a disclosure of AI-assisted writing.

But here's where it gets confusing: what if you also used ChatGPT to polish the Discussion section of that same paper? Now you need two separate disclosures:

1. Research methodology (standard Methods): "We trained a transformer model on UniProt sequences to predict binding affinity..."
2. AI writing disclosure (Methods, separate paragraph): "During preparation of this manuscript, the authors used ChatGPT (GPT-4, OpenAI) to improve the clarity of the Discussion section..."

Mixing these up creates problems. If a reviewer reads a vague statement like "AI tools were used in this study" and your paper is about an AI-based drug discovery pipeline, they'll assume you're describing your research method, not your editing process. Be explicit about the distinction.

For a paper where AI is the research subject:

"The deep learning model described in this study (ProteinForge v2.1) was developed using PyTorch and trained on curated datasets as described in Methods. Separately, during manuscript preparation, the authors used Claude (Claude 3.5, Anthropic) to improve the readability of the Introduction and Results sections. All AI-suggested text revisions were reviewed by the corresponding author, and the authors take full responsibility for the published content."

For a paper where AI isn't the research subject:

"During preparation of this manuscript, the authors used ChatGPT (GPT-4, OpenAI) to assist with editing the Methods and Discussion sections for language clarity. GitHub Copilot (Microsoft) was used to assist with writing data visualization scripts in Python. All outputs were reviewed and verified by the authors, who take full responsibility for the content of this article."

For a paper with no AI use in writing:

No disclosure needed. You don't have to state "no AI tools were used in manuscript preparation." If your paper uses AI as the research subject, that's described in Methods as methodology, it's not an AI disclosure.

A frequent question from Nature Biotechnology authors: do AlphaFold predictions, molecular dynamics visualizations, or other computational biology outputs count as "AI-generated images"?

The answer is no, with caveats.

Not covered by the image ban:

• AlphaFold or ESMFold protein structure predictions rendered as figures
• Molecular dynamics simulation snapshots
• Phylogenetic trees generated by ML-based alignment tools
• Heatmaps, volcano plots, and other standard data visualizations created with AI-assisted tools (seaborn, plotly)

These are computational outputs from scientific analysis. They represent real data processed through computational methods, which is fundamentally different from asking DALL-E to generate a figure.

Covered by the image ban:

• Using Midjourney to create a graphical abstract illustrating your research concept
• Using DALL-E to generate a schematic of a biological pathway
• Using any generative AI to create an illustration that doesn't derive from actual experimental or computational data

The line is between computational results (allowed, even if AI-assisted) and generative images (banned). If the image derives from your data, it's a result. If it was generated from a text prompt to create something visual, it falls under the ban.

Science (AAAS) initially banned all AI-generated text in January 2023 before switching to a disclosure model later that year. Nature Biotechnology never had a blanket ban, the Springer Nature approach was always disclosure-based from the start. This is a small but meaningful difference: Nature Biotechnology's policy has been more stable, which means the rules haven't changed under authors mid-submission.

The enforcement path follows standard Springer Nature procedures:

During review: Editor contacts the corresponding author and requests proper disclosure. If the reviewer specifically flagged AI-generated language, the editor may require a detailed accounting of all AI tool use before proceeding with review.

After publication: Correction, expression of concern, or retraction, depending on severity. For Nature Biotechnology specifically, a retraction of an AI/ML paper due to undisclosed AI use in manuscript preparation would be particularly damaging, it would undermine confidence in both the writing and the underlying computational work, even if the research itself is sound.

The reputational multiplier: Nature Biotechnology is a small community journal. If you publish AI/ML research and get flagged for undisclosed AI use in writing, the irony isn't lost on anyone. Researchers who build AI tools are expected to understand and follow AI disclosure policies.

For AI/ML research papers:

• Clearly separate your research methodology (the AI model you built) from your writing tools (ChatGPT, Claude, etc.) in the Methods section
• Use distinct subheadings if needed: "Machine Learning Model" for your research and "AI-Assisted Manuscript Preparation" for your writing disclosure
• If your model's architecture is described in detail, reviewers won't confuse it with a writing tool, but only if you keep the disclosures in separate paragraphs

For experimental biotech papers:

• If you used AI tools only for language editing, a standard one-paragraph disclosure in Methods is sufficient
• If AI helped with bioinformatics code (sequence alignment scripts, structural analysis pipelines), disclose this and confirm independent validation
• Make sure your code is available on GitHub or a similar repository, Nature Biotechnology expects code availability for computational methods

For reviews and perspectives:

• AI can help organize literature and structure arguments, but Nature Biotechnology's editors expect original insight that differentiates your review from a literature summary
• Disclose AI assistance in literature organization if applicable
• Don't use AI to generate the "future directions" section, that's where the editor expects your expertise, not a language model's predictions

Before submission checklist:

• [ ] Research AI and writing AI clearly separated in Methods
• [ ] AI disclosure includes tool name, version, and use case
• [ ] No generative AI images (computational outputs from real data are fine)
• [ ] All code deposited in public repository with documentation
• [ ] Co-authors aware of AI disclosure
• [ ] Computational biology figures clearly labeled as derived from data, not generated

For code and data sharing:

Nature Biotechnology expects AI/ML papers to provide:

• Source code on GitHub, GitLab, or Zenodo
• Trained model weights where feasible
• Training and test datasets (or clear descriptions if proprietary)
• Environment specifications (conda environment files, Docker containers)

This isn't part of the AI disclosure policy per se, but it's closely related. If you used AI to help write your analysis code, the code should be available so reviewers can evaluate what was AI-generated and what was human-written.

A free manuscript assessment can help you check whether your Nature Biotechnology submission meets the journal's editorial and disclosure requirements before you submit.