Guidelines by field
AI guidelines in science and academic publishing what the field's own bodies ask
Science draws its line at authorship: AI can assist, but only people can be named, and blamed. Publishing ethics bodies, journals, funders and the European Commission agree on the core, and differ mainly on how much use must be disclosed.
Last reviewed 27 October 2026. Every claim links to its source.
At a glance
AI cannot be an author
It cannot take responsibility, declare conflicts or hold copyright.
Humans stay fully accountable
For everything in a paper, including what AI wrote.
Disclose substantial use
Name the tool and say how output was checked. Spellcheck usually needs no mention; when unsure, disclose.
Keep peer review confidential
Reviewers must not upload manuscripts or grant applications to public AI tools.
No fabricated images
AI-generated or photorealistic images are restricted or banned unless editors agree.
Who has spoken
- Binding law, regulation, court ruling or enforceable professional rule
- Voluntary guidance, principles or a code an organization chooses to follow
Voluntary
Committee on Publication Ethics
Position statement: Authorship and AI tools
AI tools cannot meet the requirements for authorship; authors must disclose how they used them and remain responsible for the content.
Voluntary Jan 2026
International Committee of Medical Journal Editors
Recommendations, January 2026 update
A new section on AI, including disclosure duties for reviewers as well as authors.
Binding
Nature Portfolio
Editorial policy on AI
No AI authors, disclosure of AI use, and limits on AI-generated images. Journal policies are conditions of submission.
Binding
Science journals (AAAS)
Editorial policies on AI
Acknowledgment for AI-drafted text but not for language polishing, and restrictions on AI images.
Binding Jun 2023; Jul 2025
US National Institutes of Health
Notices on AI in peer review and AI-developed applications
Reviewers may not use AI on applications. From September 2025, applications substantially developed by AI are not treated as original, and investigators are capped at six applications a year.
Binding Dec 2023; Dec 2025
US National Science Foundation
Notice on generative AI in merit review; December 2025 revision
Limits AI in merit review; the 2025 revision names AI tools as a possible means of research misconduct.
Voluntary 4th version, Oct 2026
European Commission and ERA Forum
Living guidelines on responsible use of generative AI in research
Updated regularly; the latest adds guidance on AI agents, AI in meetings, hidden prompts planted to steer AI reviewers, and environmental footprint.
Accountability cannot be delegated to AI systems.
Binding and voluntary
Binding
Funders' rules (NIH, NSF) are conditions of funding, and breaches can lead to misconduct findings; journal policies are conditions of submission.
Voluntary
COPE and ICMJE recommendations, which member journals adopt, and the EU living guidelines.
What changed in 2025 and 2026
- The NIH stopped treating applications substantially developed by AI as original work from September 2025.
- Medical journal editors added a dedicated AI section in January 2026.
- The EU living guidelines reached their fourth version on 15 October 2026, adding AI agents, hidden prompts and environmental footprint.
Where people still disagree
Where the disclosure line sits varies: some journals ask for acknowledgment of drafting help but not language polishing, while the EU guidelines say to disclose whenever in doubt.
Questions to ask yourself
- Could I defend every sentence, figure and reference in this paper as if AI had nothing to do with it?
- Have I disclosed AI use precisely enough that a reader could judge its effect?
- Am I about to paste someone else's confidential manuscript or data into a tool?
How this lines up with my everyday guidelines
| My everyday guidelines | The field's nearest rule |
|---|---|
2. Practice transparent collaboration Own all outputs that carry your name or business reputation. | Humans stay fully accountable for everything in the paper (COPE; Nature). |
6. Stay curious and critical Check sources and verify claims, especially for health and medical information. | Say how AI output was checked for errors and bias (EU living guidelines). |
5. Protect privacy and boundaries Be mindful of confidential business or family information. | Never upload a manuscript under review to a public AI tool (NIH; ICMJE). |