Will AI review my paper? Springer Nature, Google DeepMind and the rules for AI in peer review

On 26 October 2026, Springer Nature and Google DeepMind announced a research partnership to co-design an AI assistant for peer reviewers. It is built on Google DeepMind's Gemini models and Google's open-source Agent Development Kit, and it runs inside Springer Nature's private Gemini Enterprise environment, so the content used to build and test it stays with the publisher. A prototype is being tested. Neither company has named the journals that will use it or said when it will reach them.

The announcement touches two questions researchers now ask: will an AI review my paper, and may I use AI when I review someone else's? The short answers, from the announcement and the publishers' own rules: the assistant is meant to help a human reviewer, not to write the review, and every publisher and funder we checked forbids reviewers to paste a manuscript into a public chatbot. Beyond that, the rules differ in ways that matter, and a model inside the publisher's systems may already be reading your submission.

Below: what was announced, why publishers want help, what 13 publishers, journals, funders and conferences let reviewers do with AI today, and what this means for authors. Every source was checked on 2 November 2026.

What did Springer Nature and Google DeepMind announce?#

A research partnership, not a product launch. The press release, "Springer Nature and Google DeepMind collaborate to enhance the peer review experience", says: "The assistant will help researchers to review papers more effectively, support them in navigating manuscripts, and identify areas that may need closer attention so they can spend more time on the questions that require specialist scientific judgement." It "helps reviewers analyse large amounts of text, charts, and figures."

On data, the release is specific: "The system is built using Google's open-source Agent Development Kit (ADK). It runs solely in Springer Nature's private, secure Gemini Enterprise platform. This means any content used to develop and test the AI peer reviewer assistant remains within Springer Nature's secure environment, ensuring author and reviewer confidentiality is maintained throughout." Research Information reported the same point as "manuscripts and other content used to develop and test the system".

On status: "The prototype assistant currently being tested will help re-envision peer review as an interactive experience that uses AI to empower scientific experts while keeping expert human judgment at the centre of the process." The release adds that "further updates will be shared as testing progresses", and Research Information notes that "the companies have not yet announced a timetable for wider deployment of the assistant." The release places the project in a wider effort that includes Snapp, Springer Nature's in-house peer review platform, which a March 2026 release said is used by "more than 50% of Springer Nature journals".

Saskia Steinacker, Chief Product Officer at Springer Nature, framed it around responsibility:

The expertise and judgement of reviewers are fundamental to peer review. As research grows in volume and complexity, we have an opportunity to explore how AI can responsibly support that expertise, helping reviewers focus their time where it matters most and on the questions that require specialist scientific judgement. For us, innovation and trust have to go hand in hand. That means testing what AI can add while being clear about where responsibility remains.

Seshu Ajjarapu, Senior Director, Applied AI, at Google DeepMind, framed it around volume: "Together, we are building a platform that relieves the strain of growing submission volumes while strictly upholding the editorial standards of the world's leading journals."

Research Professional News and Nature Asia's Japanese release carried the same details.

What are ScholarPeer and Deep Review?

The release says the assistant builds on "Google's experience in developing proven agentic research tools like ScholarPeer and Deep Review". It does not say how much of either goes into it. Both are public:

• ScholarPeer, described on the Google Research blog on 8 April 2026, is "a reviewer agent that automatically and rigorously evaluates academic papers": one agent grounds the review in the literature, a "baseline scout" hunts for comparisons the authors missed, and a question-and-answer engine checks the claims. Google calls it an experimental research prototype whose reviews "should not be relied upon as the definitive basis for editorial or publication decisions."
• Deep Review went public on 14 October 2026. It grew out of Google's Paper Assistant Tool, which gave "private, pre-submission feedback to authors of more than 16,000 manuscripts who opted in" at STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027. That feedback "was completely decoupled from the official conference review process". Deep Review "checks logical consistency and mathematical statements, suggests improvements, grounds facts, and compares with existing work", "while leaving higher-level judgements to humans". Anyone gets two free reviews a week.

The new assistant puts that kind of machinery next to the human reviewer, inside the publisher.

Will an AI review my paper?#

Not as the reviewer, if Springer Nature follows its own rules. Its policy on AI use in peer review sorts reviewer AI use into green, amber and red. Red, "not permitted", covers "generating full review reports; producing accept/reject recommendations; delegating critique to AI". The Nature Portfolio AI policy lists "Delegating peer review to an LLM" as red. The release describes the assistant in those terms: navigating, flagging, analyzing text and figures, with "expert human judgment at the centre".

A model may read your paper, though. At several large publishers, AI tools already help screen and check submissions. The policies and releases we read describe them:

• Springer Nature said in March 2026 that "in 2025, over 1.5m papers benefited from nearly 60 AI tools supporting screening, editorial evaluation, retention and research integrity". Its Editor Evaluation tool "was used across nearly half a million manuscripts, helping editors rapidly know if a paper is based on sound science", and its Peer Reviewer Recommender "generated over 400,000 recommendations".
• Nature Portfolio: "In the editorial process, artificial intelligence is only used to summarise, retrieve or highlight information. All decisions are made by editors."
• Elsevier editors "have access to private AI tools" for "identifying relevant reviewers, matching the scope of submissions to their journal, and detecting duplicate submissions".
• Wiley's Research Exchange platform "uses AI to support editors with reviewer matching and integrity screening".
• PLOS "uses AI tools to support submission checks, screening, and integrity investigations".
• The Science journals allow editorial offices to use AI "in a way that maintains confidentiality (i.e., closed LLMs)", and "the name of the tool and how it is used must be posted on the journal website".

What is new in the Springer Nature project is where the model sits: with the reviewer, not only in the editorial office.

Why do publishers want AI help with peer review?#

Because the volume of papers is rising faster than the supply of reviewers' time. Three examples from the past five weeks:

arXiv. In a blog post on 1 October 2026, arXiv reported 40,363 submissions in September 2026, against 20,569 in September 2024 and 9,869 in September 2016. Moderators see "an increase in thin papers of narrow scope", "salami" papers and "a marked increase in dense, AI-written papers". Since 1 October, each submitter may make "up to two submissions per calendar month, with a limit of three total active submissions at any given time". Thomas Dietterich, chair of the arXiv Editorial Advisory Council, is quoted: "a relatively small proportion of authors are submitting a large number of low-quality papers and consuming a disproportionate fraction of the moderators' time."

Scientific Reports. Retraction Watch reported on 30 October that the US National Library of Medicine (NLM) has stopped indexing new Scientific Reports articles in MEDLINE while it reviews the journal. "Following its significant growth in recent years, Scientific Reports is currently being reviewed by the US National Library of Medicine (NLM)," the journal's chief editor, Rafal Marszalek, wrote in an email on 20 October. Retraction Watch counts 28,593 Scientific Reports articles indexed in PubMed in 2026, the latest published on 15 September, while the journal's website says it has published 41,122 this year. NLM did not say why it is reviewing the journal; its policy, as Retraction Watch quotes it, re-evaluates journals if "there are significant changes to a journal", and NLM monitors "changes in publication practices, such as annual publication volume". Scientific Reports is published by Springer Nature.

AI conferences. Google's Deep Review post counts "over 74,000 submissions to three major AI conferences (ICLR, NeurIPS, and ICML)" this year, "up from approximately 45,000 in 2025, 32,000 in 2024, and 23,000 in 2023". Each needs several reviews.

Reviewers are turning to AI too, not always within the rules. Research Professional News quoted Roisin Cheshire, head of individual-led research at the funder Research Ireland, on reviews of grant applications, speaking at a European Universities Alliances conference in Brno in October: "We have rules in place at the moment around the use of AI [but] we are struggling with that a little bit. We're starting to see some reviews come in which are very obviously AI-generated. How do you police that, and how do you say to that reviewer, we can't use the review?" The same article notes that Frontiers said in December 2025 it was encouraging a "responsible, supervised" approach to AI in peer review.

The Springer Nature project offers reviewers a tool inside the publisher's own systems instead.

May I use AI when I review a paper?#

For narrow tasks, at most publishers, yes. To read the manuscript for you in a public chatbot, nowhere. Here is what each policy says, as read on 2 November 2026:

In each policy's words:

• Springer Nature: "The key question is not whether AI is used, but how it is used, and whether the peer review report remains clearly the product of independent expert judgement." In secure tools, "limited use of manuscript content may be acceptable", depending on their terms. Its advice: "Use AI for 'what should I look for?' not 'what should I conclude?'"
• Nature Portfolio: "Reviewers should disclose any AI use in their report".
• Elsevier: "Reviewers should not upload a submitted manuscript or any part of it into an AI tool". Its reviewer FAQ adds: "Even if an AI tool is described as 'private,' uploading a manuscript still means sharing confidential and unpublished research with a third-party system."
• Wiley: "Manuscripts under review are confidential and must not be uploaded to AI Technologies, whether in full or in part." "Editors have the right to disregard peer reviewer reports that show signs of inappropriate AI use."
• Taylor & Francis: "Reviewers must not use artificial intelligence tools to generate manuscript and proposal review reports, including LLM based tools like ChatGPT. However, Generative AI may be utilised to assist with improving review language."
• IEEE: "Information or content contained in or about a manuscript under review shall not be processed through a public platform (directly or indirectly) for AI generation of content for a review. Doing so is considered a breach of confidentiality because AI systems generally learn from any input."
• ACM: "Reviewers may use generative AI or other third-party tools with the sole purpose of improving the quality and readability of reviewer reports for the author, provided any and all parts of the review that would potentially identify the submission, author identities, reviewer identity, or other confidential content is removed prior to uploading into third party tools."
• Science journals: "Reviewers may not enter any part of the manuscript into an LLM or other AI tool because this could breach the confidentiality of the manuscript." Refining the review's language is allowed "if the system does not save or use inputs for model training; the name of the tool and how it was used must be declared in the review".
• PLOS: "AI tools cannot serve as peer reviewers or as decision-issuing editors, and should not be used by reviewers or editors to summarize or evaluate submission content." "Any AI use by a reviewer or editor must be disclosed".

What do funders and AI conferences allow?#

Funders are stricter, and two of the largest AI conferences have taken different routes:

• NIH: "The NIH prohibits NIH scientific peer reviewers from using natural language processors, large language models, or other generative Artificial Intelligence (AI) technologies for analyzing and formulating peer review critiques for grant applications and R&D contract proposals." NIH's peer review policy page still lists this notice; we found no newer NIH notice that changes it.
• NSF: "NSF reviewers are prohibited from uploading any content from proposals, review information and related records to non-approved generative AI tools." Reviewers "may share publicly available information with current generation generative AI tools."
• NeurIPS 2026: the Main Track Handbook says: "Do not share, discuss, or disclose any information related to submissions with anyone, including any unsanctioned LLMs." It describes a planned experiment, "though the details may change", in which every reviewer has some papers where they may use "LLM support that we will provide via OpenReview" and some where no LLM is allowed.
• ICLR 2027: its reviewer policy permits "limited and responsible use of AI tools to assist human writing of peer reviews", but "it is mandatory to disclose use of AI tools". If you use an LLM "to generate or edit any portion" of a review, you must report "your original, self-written paper assessment, and any LM interactions". LLM-generated falsehoods or hallucinations can lead to "desk-rejecting all papers authored by the (meta) reviewer".

Where do the rules agree, and where do they split?#

They agree on three things. The reviewer stays responsible for the review. AI does not write the review or make the recommendation. Unpublished work stays confidential.

They split on tools. Elsevier, Wiley, Taylor & Francis, the Science journals, PLOS and NIH keep manuscript content out of AI tools altogether; Elsevier and the Science journals also limit the tools a reviewer may use on their own text to private ones, or ones that do not train on inputs. Springer Nature, IEEE, ACM, NSF and NeurIPS draw the line elsewhere: at public, unsecured, unapproved or unsanctioned tools. That leaves room for a secure tool that may see the manuscript.

A publisher-run assistant sits in that room: a secure tool the publisher provides itself. Wiley lists "confidential review assistance" within "secure editorial environments" among possible future uses, and NeurIPS gives some reviewers a sanctioned model through OpenReview. None of the policies we read says yet how a reviewer should declare help from a publisher's own assistant.

What does this mean for authors?#

A model may read your manuscript, inside the publisher. That is within the publishers' own rules, and the same rules forbid your reviewers from pasting it into a public chatbot. If you want to know which tools a journal uses, ask it; the Science journals' guidelines say editorial offices must post theirs on the journal website.

Expect your AI disclosure to be read closely. Wiley tells its editors: "AI use in manuscript development is not grounds for rejection." The questions it asks instead include: "Are the disclosures regarding AI use consistent with what's presented in the manuscript?" PLOS says it "may use available AI detection tools in article-level checks or assessments", and that undisclosed AI use may prompt an integrity investigation. Springer Nature says its AI tools identified 25,000 papers in 2025 with "issues such as image manipulation, fake references and fabricated text". A clear statement of what AI did, and a reference list you have checked, answer those questions before they are asked. Our guides cover how to disclose AI use and how to check for hallucinated references, and Papex's journal AI policies page quotes what 29 publishers, journals, preprint servers and guidelines ask authors to say. The same week's record retraction at IEEE shows which checks caught paper-mill and AI-faked papers: Why IEEE retracted 1,400 papers in one day.

The rules reach into your response to reviewers. The Science journals' September 2026 guidelines say that entering "all or part of the review into an LLM or other AI tool to generate or augment a response" is not allowed; refining the language of your response is allowed, with the tool declared.

What is still unknown about the assistant?#

The announcement leaves open what most researchers will want to know:

• Which Springer Nature journals will test or use it, and when.
• Whether reviewers choose to use it, and whether authors are told.
• Whether reviewers must declare its use, as Springer Nature's policy asks them to declare other AI use.
• Whether editors or authors see what it flagged.
• How well it works. We found no published evaluation of the prototype.

What can a researcher do today?#

As a reviewer:

1. Read the journal's reviewer policy first. Journals within a publisher can set stricter rules than the tables above.
2. Keep the manuscript out of public AI tools. Work from your own notes, as Springer Nature suggests, and ask general questions.
3. Write your own assessment before any AI polishing. ICLR now requires that draft; it is good practice anywhere.
4. Declare what the tool did, in the form the journal asks.

As an author, record AI use while you work, not afterwards: which model and version, what it did, and who checked the result.

This is the part Papex works on: a signed record of what AI did in one version of your work.

• While you work. An AI app connected to Papex can prepare private drafts and leave a private summary of each working session. It "cannot sign, accept its own proposals, make rights statements, publish, or change a published record", and "an AI app is never listed as an author" (guide to AI apps).
• When you sign. An "AI in this version" card shows what connected apps reported doing, with AI work you declare yourself. Signing confirms it, and the signed version keeps it under "Who did what". Model and tool names are "kept as the AI agent reports them and marked Agent-reported"; Papex does not verify them (FAQ).
• For the journal. From the signed record, Papex writes the AI-use statement in one of the forms Journal AI policies lists: Elsevier, Springer Nature, Wiley, Taylor & Francis, IEEE, ACM, arXiv and AMS, or a general ICMJE and COPE form.

A signature links the people who signed to one exact version, and its timestamp shows that the version existed by that date. Neither establishes scientific validity or peer review, and no publisher in the tables asks for a Papex record: you paste the statement where your journal asks for it.

Sources#

• Springer Nature and Google DeepMind collaborate to enhance the peer review experience (Springer Nature press release, 26 October 2026), checked 2 November 2026
• シュプリンガーネイチャーとGoogle DeepMind、査読体験の向上に向けて提携 (Nature Asia, Japanese press release, 26 October 2026), checked 2 November 2026
• Springer Nature and Google DeepMind partner on AI peer review assistant (Research Information, 26 October 2026), checked 2 November 2026
• Rachael Pells, Springer Nature and Google DeepMind to build 'peer review assistant' (Research Professional News, 26 October 2026), checked 2 November 2026
• Springer Nature embraces AI tools across the publishing process, resulting in less friction and increased author satisfaction (Springer Nature press release, 12 March 2026), checked 2 November 2026
• Jinsung Yoon and Tomas Pfister, Improving the academic workflow: Introducing two AI agents for better figures and peer review (Google Research blog, 8 April 2026), checked 2 November 2026
• Rajesh Jayaram and Vincent Cohen-Addad, Analyzing technical documents with Deep Review (Google Research blog, 14 October 2026), checked 2 November 2026
• Kat Boboris, Fair Moderation, Equitable Access, and AI: arXiv's Updated Rate Limit Policy (arXiv blog, 1 October 2026), checked 2 November 2026
• Kate Travis, Scientific Reports on hold in PubMed while NLM reevaluates journal (Retraction Watch, 30 October 2026), checked 2 November 2026
• AI Use in Peer Review (Springer Nature editorial policies), checked 2 November 2026
• Artificial Intelligence (AI) (Nature Portfolio editorial policies), checked 2 November 2026
• Generative AI policies for journals (Elsevier, updated June 2026), checked 2 November 2026; the older address for Elsevier's review-process policy now redirects to its policy for books
• Ethics guidelines: AI section (Wiley, last updated July 2026) and AI guidelines for researchers (Wiley), checked 2 November 2026
• AI Policy (Taylor & Francis), checked 2 November 2026
• Submission and peer review policies: Guidelines for AI-generated content (IEEE Author Center), checked 2 November 2026
• ACM Peer Review Policy (ACM), checked 2 November 2026 through the Internet Archive copy of 13 May 2026; the live page blocked automated access
• Science Journals: Editorial Policies (AAAS), checked 2 November 2026 through the Internet Archive copy of 14 June 2026, and Science journals' guidelines for AI usage (September 2026 PDF), read through the Internet Archive copy of 27 October 2026; the live pages blocked automated access
• Ethical Publishing Practice: Artificial Intelligence Tools and Technologies, Ethical Peer Review and Confidentiality (PLOS ONE), checked 2 November 2026
• NOT-OD-23-149: The Use of Generative Artificial Intelligence Technologies is Prohibited for the NIH Peer Review Process (NIH, 23 June 2023), checked 2 November 2026 through the Internet Archive copy of 24 October 2026, with the NIH peer review policies page (Internet Archive copy of 2 October 2026); the live pages blocked automated access
• Notice to Research Community: Use of Generative Artificial Intelligence Technology in the NSF Merit Review Process (NSF, 14 December 2023, last updated 21 May 2026), checked 2 November 2026
• Main Track Handbook (NeurIPS 2026), checked 2 November 2026
• Policy on LLM use for writing reviews or meta reviews (ICLR 2027), checked 2 November 2026