Can you use AI for peer review? What NIH, NSF, journals, and conferences allow

The Morse teamPublished October 25, 20269 min read

You can use AI for peer review only around the edges, and for an NIH grant review not at all. As of October 2026, NIH bans its reviewers from using generative AI for critiques, NSF bans uploading any proposal content to tools it hasn't approved, and Elsevier, Springer Nature, Wiley, Science, and PLOS all forbid putting a submitted manuscript into a public AI tool. Several publishers do let you polish the wording of your own review, but some also set a test for the tool. Consumer ChatGPT, Gemini, and Claude accounts, which save your chats on the company's servers, fail at least one of those tests.

We read NIH's and NSF's notices, the ICMJE Recommendations, five publishers' AI policies, and the reviewer rules for NeurIPS 2026, ICLR 2027, and ACL Rolling Review on October 25, 2026. Elsevier and Springer Nature deserve credit for explaining, in terms a reviewer can check, what makes an AI tool safe for confidential work. Your journal's or funder's own rule beats this page.

Can you use ChatGPT for peer review?

Not with the manuscript in it. The ICMJE Recommendations, updated January 2026, say reviewers "must adhere to the journal's stated policy on the use of AI or request permission from the journal" first, that confidentiality "may prohibit the uploading of the manuscript" to AI tools "where confidentiality cannot be assured unless such use of AI is explicitly permitted," and that reviewers who use AI "should disclose its use" (ICMJE).

Polishing your own review is the use these policies allow most often, and several attach a condition to the tool:

• Science journals let reviewers "use AI tools to revise their own writing, provided that the system does not save or use inputs to train the model," and the use must be declared in the review (Science, archived June 14, 2026).
• Elsevier, in a policy updated June 2026, says "reviewers should only use private AI tools," meaning a version that "does not retain, re-use, share, or learn from the content you submit to it," and tells reviewers to "prefer paid or enterprise versions over free tools" (Elsevier).
• ACL Rolling Review treats the review itself as confidential: reviewers may not upload "their peer review report into a non-privacy-preserving generative tool, even if it is just for the purpose of improving language and readability" (ACL Rolling Review).

Consumer accounts fail both Science's test and Elsevier's. OpenAI says that on ChatGPT Free, Plus, and Pro, "we may use your content to train our models" unless you turn off Improve the model for everyone (OpenAI, archived October 19, 2026). Google trains on Gemini chats while Keep Activity is on and keeps them 72 hours even with it off (Google). Anthropic may keep Claude chats "for up to 5 years" in de-identified form if model improvement is on (Anthropic). We cover the ChatGPT side in does ChatGPT train on your data.

What does NIH's AI peer review policy say?

It bans the use itself. NOT-OD-23-149, released June 23, 2023, "prohibits NIH scientific peer reviewers from using natural language processors, large language models, or other generative Artificial Intelligence (AI) technologies" for critiques, and says that uploading or sharing "content or original concepts from an NIH grant application" to online AI tools violates NIH's confidentiality rules. Reviewers certify this when they sign a modified Security, Confidentiality and Nondisclosure Agreement before each meeting, and the rule covers National Advisory Council members too. NIH explains that "AI tools have no guarantee of where data are being sent, saved, viewed, or used in the future" (NIH).

The consequences come from NOT-OD-22-044, from December 30, 2021: ending your review service, a referral for "government-wide suspension or debarment," and possible referral to HHS's Inspector General and the Justice Department, "which may include pursuing criminal and civil penalties" (NIH). We found no later NIH notice that loosens the ban as of October 25, 2026.

Because the ban covers any generative AI, a tool that promises to keep nothing doesn't make it allowed.

Can you use AI to write an NIH grant application?

Yes, within limits that tightened in 2025. NOT-OD-25-132, released July 17, 2025 and effective from the September 25, 2025 receipt date, says "NIH will not consider applications that are either substantially developed by AI" or that contain sections substantially developed by AI to be the applicant's original ideas. NIH says it will "employ the latest technology in detection of AI-generated content," may refer a case found after award to the Office of Research Integrity, and now accepts only six new, renewal, resubmission, or revision applications per principal investigator per calendar year (NIH).

Does NSF allow AI in merit review?

Only with public information. NSF's notice of December 14, 2023, last updated May 21, 2026, says "NSF reviewers are prohibited from uploading any content from proposals" or review records to non-approved generative AI tools, and that anything entered into tools outside NSF's firewall "is considered to be entering the public domain." The page names no approved tool. Proposers, for their part, are encouraged to say in the project description "the extent to which" they used generative AI (NSF).

Who allows what: the AI rules for reviewers and authors

Read October 25, 2026, from each body's own page linked above or in the first column.

Elsevier and NeurIPS built their own tools rather than trust outside ones. Elsevier's editors get "private AI tools" inside its editorial systems. NeurIPS 2026 runs a voluntary LLM experiment inside OpenReview, for papers whose authors opted in, with LLMs that run "with zero data retention"; outside it, "NeurIPS does not sanction any other use of LLMs during the review process."

Is it safe to put unpublished research in ChatGPT?

Journals allow it, and the risk sits in the app's terms. Every publisher above lets authors use AI on their own manuscript with disclosure. Wiley's guide for researchers warns that drafting with AI can cause "unintended exposure of your unpublished findings" and tells authors to read a tool's terms for "ownership," "data reuse," "training," and "opt out" (Wiley). Springer Nature adds that some institutional AI detectors "can retain and train on content uploaded to them."

Can you put participant or IRB data into an AI tool?

Only if your IRB approval and consent forms cover it. Wiley tells researchers to "consult your Institutional Review Board (IRB)" about AI and participant data, and to check "whether your original participant consent forms cover sharing data with third-party AI services." Lehigh University's IRB guidance says to de-identify data first, warns that "file names, geolocation, timestamps, or other metadata" can disclose identity and that AI "may infer personal information from patterns in the de-identified data," and requires consent forms to name the AI tools used (Lehigh).

Controlled-access data has its own rule. NIH's NOT-OD-25-081, released March 28, 2025, says "sharing controlled-access data with public generative AI tools (e.g., third party tools) via prompts" breaks the Genomic Data Sharing Policy's non-transferability term and the Data Use Certification (NIH). This isn't legal or compliance advice; your IRB and your data use agreement decide.

How Morse keeps your own unpublished work private

The publishers' tests come down to whether the tool keeps your text and whether it trains on it. Consumer apps keep your chat history on their servers, and can train on it depending on a setting you have to find. Morse is built not to keep that copy. Your chat history is stored only in your browser, encrypted. Morse's servers don't keep messages or log message content, and Morse doesn't train AI on your content.

That doesn't change anyone's rules. Where a funder or publisher bans putting a confidential manuscript or application into any AI tool, as NIH, PLOS, Wiley, Elsevier, and Science do, the ban includes Morse: don't paste someone else's manuscript, grant application, or critique into it. If a journal allows a privacy-preserving tool for polishing your own comments, ask the editor whether Morse meets its test; that call is the journal's. Morse is for your own work: your drafts, your grant narrative, your own results, and uses your journal permits. Pick a mode per chat:

• Private, for drafts of your own paper or grant. It's the default for a new chat. Open-weight models like DeepSeek, Kimi, and GLM run under the zero data retention agreements Morse has with its inference, search, and other providers, so they've agreed to keep nothing, and where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't (what zero data retention means).
• E2EE, for unpublished results no one else should read. Your browser checks the secure enclave's proof that it's genuine, then encrypts the message so only that enclave can read it, and Morse's servers relay text they can't read. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed. E2EE has no web search, so run your literature search in another chat. It comes with Pro, at $19 a month.
• Anonymous, for general methods questions to GPT, Claude, or Gemini. Morse's proxy strips your account, email, and IP address, so the provider sees the message without knowing it's you. It may keep it for a limited time to check for abuse, under its own policy, so ask about a statistical test in general terms and leave your findings out.

In every mode, Morse doesn't sell your data or show ads. Your history locks when you sign out, and unlocking it takes a key that never leaves your device plus one our server gives you at sign-in. Because we don't store chats, we usually have very little to give authorities, as our privacy policy explains. Clearing your browser or deleting your account deletes your chats, and Morse can't recover them, so keep drafts and data where your lab keeps them. If your IRB protocol or consent doesn't cover an outside AI service, it doesn't cover Morse.

So can you use AI for peer review?

• NIH grant reviewer: no AI for the critique, in any tool, Morse included.
• NSF reviewer: nothing from the proposal goes into a non-approved tool; public information only.
• Journal reviewer: follow the journal, never paste the manuscript, and polish only your own words, in a tool that passes the journal's test, with disclosure.
• Author: disclose as each journal asks. For drafts, a Morse chat in Private or E2EE keeps your history in your browser instead of on a company's servers.
• Researcher with participant data: your IRB and consent forms first, then the tool.

You can try a chat on the Morse home page without an account, or start free and move to Pro when your unpublished results need E2EE.