Guide12 November 202618 min read25 sources

Six probability journals paused submissions over AI: the new IMS, Bernoulli and MSP rules

Why EJP, ECP, AIHP, Bernoulli, ESAIM: PS and SPA paused submissions in October 2026, which have reopened, and what the new AI rules ask of authors and referees.

Papex

On 19 October 2026, six journals in probability and statistics stopped new papers at once: Electronic Journal of Probability (EJP), Electronic Communications in Probability (ECP), the Annales de l'Institut Henri Poincaré, Probabilités et Statistiques (AIHP), Bernoulli, ESAIM: Probability and Statistics (ESAIM: PS) and Stochastic Processes and their Applications (SPA). Five closed to new submissions; SPA deferred handling them. In a joint letter, the editors gave two reasons: a sharp increase in submissions, and the rapid development of AI, which they said requires them "to rethink some of our editorial policies and tools".¹ The new rules are now public. The Institute of Mathematical Statistics (IMS) put an AI policy into effect on 8 November, Bernoulli adopted its own from the same day, and Mathematical Sciences Publishers (MSP) posted a draft on 1 November.

Some things are not established. No journal published how many papers it received, or how many involved AI. The joint letter names the increase and AI side by side without saying how much of one comes from the other; only ESAIM: PS's own notice says the increase was "driven largely" by AI.² On 12 November, three of the six journals' pages still said they were paused, and we could not read the sites of two others. MSP's text is a draft.

Below: which journals paused and why, what happened to papers under review, which have reopened, what the IMS, Bernoulli and MSP texts require of authors, referees and editors, a checklist for authors built from the policies' own words, a comparison with other mathematics publishers and arXiv, and what is still open. Every source was checked on 12 November 2026.

Which six journals paused submissions, and why?#

The announcement went out on 19 October as a letter from the editors-in-chief, posted that day by the Société Mathématique de France (SMF).¹ It reads: "We are writing to inform you that the following journals will temporarily close submissions starting October 19". Five journals closed. For the sixth it says: "The journal Stochastic Processes and Applications will be temporarily deferring the handling of new submissions received during the same period." The letter promised that submissions "will reopen during November under updated policies, which will be clearly communicated on the journals’ websites."

The editors-in-chief of all six signed it; their names are in the table further down. Under the heading "Why are we closing?", the letter gives two reasons. The first is workload: "The sharp increase in submissions has put unprecedented pressure on our editorial boards and on the entire editorial system. At the current pace, we cannot give every manuscript the care and attention we believe it deserves."¹ The second is AI: "At the same time, the rapid development of AI is profoundly changing scientific publishing and requires us to rethink some of our editorial policies and tools." The editors would use the closure to "introduce immediate measures and accelerate broader changes".

The journals' own notices differ in emphasis. EDP Sciences, which publishes ESAIM: PS, announced a "temporary suspension of new manuscript submissions from 19 October to 8 November 2026" in response to "a substantial increase in manuscript submissions across the discipline, driven largely by the rapid development of artificial intelligence."² The Bernoulli Society posted on 21 October that the journal "is temporarily pausing submissions until November 8 while the editorial board and publishers develop new editorial policies, guidelines, and practical solutions to handle the substantial increase in submissions."³ The notice on the IMS pages of EJP, ECP and AIHP uses the same sentence, with "through mid-November" in place of the date.

So the stated reason is a volume of submissions large enough to strain refereeing, with AI named as the change behind it. The Proofs and Prompts newsfeed¹⁴ summed it up on 20 October as "an influx of papers, largely due to the use of AI tools". That is the newsfeed's summary; among the journals' own words, the "largely" is ESAIM: PS's.

One phrase travelled with the story and belongs to none of them. Retraction Watch's Weekend Reads of 24 October¹⁵ wrote that the journals were closing "to cope with the Mathpocalypse / Slopwave", quoting a LinkedIn post by Oliver Johnson¹⁶, who added "(edit terminology according to taste and priors)". The words are a commentator's joke; the editors did not use them.

What happened to papers already under review?#

Only ESAIM: PS's notice says: "The suspension applies only to new submissions. Authors will still be able to submit revisions, and all ongoing editorial and peer review processes will continue as normal."² The joint letter speaks only of closing submissions, the Bernoulli and IMS notices say "New submissions will not be possible during this period", and SPA's deferral covers new submissions "received during the same period". None of them says that papers already in review were stopped or sent back.

What else was happening around the pause?#

The letter names no company or event. Two things from the weeks before set the scene.

On 1 October, arXiv limited every submitter to two new submissions per calendar month, with at most three active at a time, "as a stopgap while we determine what may be the new best practice for authors employing more and more advanced AI tools".¹⁷ It received 40,363 submissions in September 2026, against 20,569 in September 2024, and its moderators were seeing "a marked increase in dense, AI-written papers."

On 6 October, OpenAI posted a repository of mathematics produced by an internal model. Its README at the first commit said: "The current catalogue contains 722 manuscripts organized into 372 families."¹⁹ Our post on Terence Tao's warning covers that release and the reactions, and our explainer on the CMSA report covers the late-October call for a methodologies section in AI-assisted papers.

None of the six journals' notices mentions the OpenAI release. It is context here, and the journals do not give it as a cause.

Have the journals reopened?#

Not all of them, as far as their public pages show. We read each journal's page on 12 November. We could not test the submission systems, which need an account.

Journal	Signed for it	The pause as announced	Its page on 12 November 2026
EJP	Cristina Toninelli	Closed from 19 October	IMS page still opens: "temporarily pausing submissions through mid-November"
ECP	Patrícia Gonçalves	Closed from 19 October	Same notice as EJP
AIHP	Nicolas Curien	Closed from 19 October	Same notice as EJP
Bernoulli	Kengo Kato	Closed "until November 8"	No pause notice; new AI policy "applies to submissions on or after November 8, 2026"
ESAIM: PS	Benoîte de Saporta, Céline Duval	Closed 19 October to 8 November; EDP said it would reopen 9 November	Journal site refused automated access; not checked
SPA	Arnaud Guillin, Eva Löcherbach	Did not close; deferred handling of new submissions	Journal pages on Elsevier's ScienceDirect refused automated access; not checked

Table 1 The six journals, who signed the joint letter for each, and what their pages said on 12 November 2026.

The IMS pages of EJP, ECP and AIHP carry the pause notice at the top and give no date beyond "mid-November".^9,10,11 Bernoulli is the clearest case: its home page, last updated on 11 November, shows no pause and says the journal "has updated the Aims and Scope and adopted a new AI policy for authors, reviewers, and editors."⁸ For ESAIM: PS, EDP gave 9 November as the reopening date.² For SPA, the joint letter is the only text we could read, so check its own pages before submitting.

What does the new IMS AI policy require?#

The IMS policy opens: "This policy takes effect from 8 November 2026. The guiding principles for all IMS journals in the use of AI are transparency and responsibility."⁴ It has a part for authors and a part for reviewers. The IMS website links the same page from the menus of its own journals, such as the Annals of Probability and the Annals of Statistics, and from co-sponsored and supported journals, among them EJP, ECP and AIHP.

For authors. Eight rules:

Writing: "All submissions must meet very high standards of exposition and readability. AI may be used to aid in preparing the text, but should not be used for writing papers or entire sections of papers."
Research use and the statement: AI in the research "is permitted, but must be declared both in a ‘Statement on AI use’ section to appear immediately before the bibliography of the article, and in EJMS, where authors will be asked to confirm that their article complies with this policy." EJMS is the online submission system that IMS journals share.⁷ The policy adds: "The declaration should be made even if no AI was used, and otherwise should give a full and detailed description of the purposes for which it was used. Authors retain full responsibility for all aspects of their submission."
Plagiarism: "Any plagiarism that occurs due to the failure of AI to attribute sources or ideas appropriately is considered as plagiarism committed by the authors."
Authorship: "Only humans may be listed as authors."
Hidden instructions: "Authors must not include concealed text, instructions, prompts, or other material intended to manipulate reviewers, AI systems or the editorial process."
Sanctions: breaking these rules "is considered grounds for rejection of the article, and may at the discretion of the editors lead to a temporary ban on future submissions to IMS journals."
Preprints: "Authors are strongly encouraged to make their paper publicly accessible, e.g. on arXiv, at the time of submission."
Consent to AI in review: "As a condition of submission, authors agree that AI tools may be used in the review process, subject to the safeguards outlined below."

For reviewers. Reviewers "are expected to use their own judgement", and AI "should not be used for writing the report or entire sections of the report." AI tools in review are allowed, but "any use of AI by a reviewer must be disclosed to the authors in their report." On confidentiality: "The uploading of proprietary data or manuscripts that are not publicly accessible to AI tools where conversations may be used for training is strictly prohibited." Confidential editorial correspondence, including review reports, "must not be disclosed to AI systems." And: "Authors who use AI tools in line with the policy above should not be penalised in the review process."

Two points deserve a close reading. The upload ban names manuscripts "that are not publicly accessible"; the policy does not say how it treats a paper already on arXiv, though it encourages authors to post one. And there is no section for editors, who appear only as the people who may impose a ban.

What does Bernoulli's new AI policy say?#

Bernoulli's policy applies "to submissions on or after November 8, 2026" and "will be revisited and subject to change as needed".⁵ Its main points:

Authors may use AI and remain "fully responsible for the content of their submissions, including correctness, originality, presentation, and intellectual integrity." They "must carefully verify any AI-assisted material, fully understand the work they submit, and ensure that their manuscripts are comprehensible to human readers."
A statement goes in the acknowledgments. It "should explain whether and how AI tools were used in proof development, writing and exposition, and other supporting tasks, such as numerical experiments and proofreading." It is required even when no AI was used. The acknowledgments do not count toward the 25-page limit in the journal's template.⁷
The board decides. "The editorial board has final discretion to determine whether the nature and extent of AI use are acceptable." Severe violations may restrict future submissions.
Reviewers and editors may use AI for technical checks, "such as examining proofs, verifying references, and checking internal consistency", but judgements of novelty, importance and exposition "must be grounded in their own expertise." Reviewers "must specify, in their confidential comments to the editors, which AI tools they used and for what purposes."
In review, only secure tools may see a manuscript: "secure or institutionally approved AI tools whose data-control settings adequately prohibit the use of manuscript content for model training."

Bernoulli updated its Aims and Scope at the same time. The current text says: "Mere technical execution without clear motivation, without new understanding, or without a transferable idea would not be suitable for the journal."⁶

Two differences from the IMS text stand out. A Bernoulli reviewer tells the editors about AI use in confidence; an IMS reviewer tells the authors in the report. And Bernoulli's statement goes in the acknowledgments, the IMS one in its own section before the bibliography.

What does MSP's draft AI policy say?#

MSP, a nonprofit publisher whose journals include Geometry & Topology, Algebra & Number Theory, Analysis & PDE, the Pacific Journal of Mathematics and Probability and Mathematical Physics,¹³ posted its text on 1 November 2026.¹² It calls it "our current thinking", "meant as a starting point for dialogue and consultations, and a general framework for our journals that is, for now, deliberately minimal." It adds: "Our editorial boards continue to discuss, develop their policies, and adapt to the ongoing dramatic changes to the profession."

MSP "will permit their use in preparing submissions", meaning AI tools, on three terms:

Exposition counts. "Authors should keep in mind that the purpose of mathematical publication is not only to record progress, but also to communicate it to others." A paper must let "a knowledgeable reader" follow and verify its arguments, and "Submissions will be judged on the quality of their exposition as well as on the correctness and significance of their results."
Acknowledge AI beyond language help, by location. "Authors must acknowledge any use of AI beyond routine help with language, such as correcting spelling and grammar or improving the fluency of the authors’ own text." The acknowledgment "must identify which results, proofs, computations, or passages were produced or materially shaped with AI assistance".
Authors carry the responsibility. "Whatever tools are used, the authors alone are responsible for the mathematical correctness of the submission and for any claims it makes of conceptual progress". And: "The use of AI does not diminish this responsibility in any way."

The draft says nothing about referees or editors, and it does not require a statement when no AI was used.

What must an author now put in a submission?#

The three texts agree on most of what an author must do, and differ on where to put it. This list applies to EJP and ECP (IMS policy), Bernoulli and MSP's journals; for ESAIM: PS and SPA, check the journal first.

Step 7 / 7

Find the policy that applies. IMS: in effect since 8 November. Bernoulli: submissions from 8 November. MSP: a draft of 1 November.
Write an AI statement, even if you used no AI. IMS and Bernoulli both require one; MSP's draft asks only when AI went beyond routine language help.
Put it where the journal asks. IMS: a "Statement on AI use" section immediately before the bibliography, plus a confirmation in EJMS. Bernoulli: the acknowledgments.
Say what the AI did, part by part. IMS wants the purposes in full detail; Bernoulli names "proof development, writing and exposition, and other supporting tasks"; MSP asks which "results, proofs, computations, or passages" AI shaped.
Write and check the paper yourselves. At IMS journals, AI may help with the text but not write the paper or whole sections. Check every reference: IMS counts plagiarism caused by AI as the authors' own.
List only people, and hide nothing. Humans only as authors (IMS and Bernoulli); no concealed text or prompts (IMS).
Expect AI in review, and consider a preprint. IMS and Bernoulli make AI-assisted review, under safeguards, a condition of submission. Both encourage a public preprint: IMS names arXiv, Bernoulli arXiv and HAL. arXiv's limit of two new submissions a month applies.

Fig. 1 What an author submitting to an IMS journal, Bernoulli or an MSP journal now needs to do, from the policies' own words.

The policies give no template for "full and detailed". Our guide to writing an AI disclosure statement has worked examples, and the guide to reporting the AI model version covers model names and dates.

How do other mathematics publishers and arXiv compare?#

We read each publisher's own page on 12 November. The table keeps to what those pages say.

Venue (date)	What authors must state	Limits on AI	AI in refereeing
IMS journals⁴ (in effect 8 November 2026)	"Statement on AI use" before the bibliography, also when no AI was used; confirmed in EJMS	Not for "writing papers or entire sections of papers"; humans only as authors; no concealed prompts	Allowed; disclosed to the authors in the report; no non-public manuscripts in tools that may train on them
Bernoulli⁵ (submissions from 8 November 2026)	Statement in the acknowledgments, also when no AI was used	Humans only as authors; the board decides if use is acceptable	Technical checks allowed; tools named to editors in confidence; secure tools only
MSP¹² (draft, 1 November 2026)	Any use "beyond routine help with language", with the results, proofs, computations or passages it shaped	Authors "alone are responsible"	Not addressed
Annals of Mathematics²⁰	"All manuscripts must include an AI USE DECLARATION"; an idea or proof from AI described and located	"AI agents cannot be named authors"	Not addressed on the page
London Mathematical Society²¹ (updated October 2026)	An "AI Declaration" section in all manuscripts, with "the name and model number of the AI tool"	Generative AI cannot be an author	Not for reviewing; uploading a manuscript to an AI tool is "misconduct"
Elsevier²² (all its journals)	A separate AI declaration; basic grammar and spelling checks excluded	AI not an author; rules for AI in figures	Reviewers "should not upload a submitted manuscript or any part of it into an AI tool"
Springer Nature²³ (all its journals)	Describe low-risk (green) and amber uses	Not permitted: "Writing results or discussion independently; generating scientific claims or interpretations"	Not permitted: uploads to public tools, full review reports written by AI
arXiv¹⁸ (moderation policy)	Report "significant use" of tools, including text-to-text generative AI	AI not an author; two new submissions a month, three active	Moderation "is not a peer-review process"

Table 2 What mathematics publishers and arXiv say about AI on their own pages, as read on 12 November 2026.

Behind the table: the LMS counts as AI use "generating ideas, performing calculations, constructing or verifying proofs, improving language, or in any other manner", and lets reviewers use AI only to polish their own report text, declared.²¹ Springer Nature applies "a risk-assessment approach, focusing on how AI is used, rather than whether it is used";²³ its peer review page puts "Uploading manuscripts to public/unsecured tools" among the uses not permitted.²⁴ arXiv holds authors responsible for content "irrespective of how the contents were generated".¹⁸

Three publishers are missing. The American Mathematical Society's page on AI use and SIAM's information for authors refused our automated requests on 12 November, and EMS Press's information for authors sends authors to each journal's page,²⁵ which we did not check one by one. We left them out rather than describe them from memory.

Side by side, the new probability rules go further than most. IMS and Bernoulli, like the Annals and the LMS, want a statement even when no AI was used. IMS is the strictest here on writing: AI may help prepare the text but not write a paper or a whole section. And only IMS and Bernoulli make AI-assisted refereeing a condition of submission, while the LMS and Elsevier bar reviewers from uploading manuscripts to AI tools.

What is still undecided?#

Much of this is new, and some of it is provisional.

Reopening dates. On 12 November, EJP, ECP and AIHP still said "through mid-November"; ESAIM: PS and SPA could not be checked.
ESAIM: PS and SPA policies. EDP said ESAIM: PS would review "its aims and scope and guidance on the use of AI by authors and reviewers".² We could not read the result.
The "practical solutions" promised for the volume of submissions. Bernoulli's new scope is visible; we found no journal that has published a cap on submissions or a new triage rule.
MSP's final text. MSP calls its framework "deliberately minimal", and its editorial boards "continue to discuss, develop their policies".
Bernoulli's next version. Its policy "will be revisited and subject to change as needed".
How the rules are applied. "Full and detailed" (IMS) and "nature and extent" (Bernoulli) leave room for judgement, and sanctions are at the editors' discretion.
Numbers. No journal has said how many submissions it received, rejected, or found to involve AI.

What this means for a research record#

The new rules ask authors to say which results, proofs, computations or passages AI shaped, and to stand behind every part. That is easiest to answer from notes kept while working: which tool and model, as the app or API named it; which tasks it did; which sections or lemmas came out of that work, and who checked them; the dates; and which version of the manuscript went to which journal on which day.

On Papex, the author can sign one exact version of a paper with its AI-use statement, and the signed version carries its date. A signed record shows who signed which version, when, and what they said the tools did. It is not peer review and does not show that the mathematics is right, and an AI is never listed as an author.

Sources#

Fermeture temporaire de soumissions: joint letter from the editors-in-chief of AIHP, Bernoulli, ECP, EJP, ESAIM: P&S and SPA (Société Mathématique de France, 19 October 2026), checked 12 November 2026 a b c
ESAIM: Probability and Statistics takes action to address increasing manuscript submissions (EDP Sciences, 19 October 2026), checked 12 November 2026 a b c d e
News: Temporary Submission Pause (21 October 2026) and A Message from the Editor of the Bernoulli Journal (5 November 2026) (Bernoulli Society), checked 12 November 2026 ↩
Policy on Use of AI, in effect from 8 November 2026 (Institute of Mathematical Statistics), checked 12 November 2026 (read in a browser; the site blocks simple automated requests) a b
Bernoulli Journal: Policy on AI (Bernoulli Society, last updated 5 November 2026), checked 12 November 2026 a b
Bernoulli Journal: Aims and Scope (Bernoulli Society, last updated 5 November 2026), checked 12 November 2026 ↩
Bernoulli Journal: Notes for Authors (Bernoulli Society), checked 12 November 2026 a b
Bernoulli Journal home page (Bernoulli Society, last updated 11 November 2026), checked 12 November 2026 ↩
Electronic Journal of Probability (Institute of Mathematical Statistics), checked 12 November 2026 ↩
Electronic Communications in Probability (Institute of Mathematical Statistics), checked 12 November 2026 ↩
Annales de l'Institut Henri Poincaré, Probabilités et Statistiques (Institute of Mathematical Statistics), checked 12 November 2026 ↩
Artificial Intelligence: Draft Policy on the Use of Artificial Intelligence (MSP, 1 November 2026), checked 12 November 2026 a b
Our journals (MSP), checked 12 November 2026 ↩
Newsfeed, item of 20 October 2026 (Proofs and Prompts), checked 12 November 2026 ↩
Weekend reads: Journals and the 'Mathpocalypse'; autism paper earns EOC; Cell Press investigating bribery allegations (Retraction Watch, 24 October 2026), checked 12 November 2026 ↩
Oliver Johnson, post on probability journals closing to submissions (LinkedIn, October 2026), checked 12 November 2026 ↩
Fair Moderation, Equitable Access, and AI: arXiv's Updated Rate Limit Policy (arXiv blog, 1 October 2026), checked 12 November 2026 ↩
arXiv moderation, including the policy for authors' use of generative AI language tools and the submission rate (arXiv), checked 12 November 2026 a b
openai/math README at the initial commit adc7f12 (GitHub, 6 October 2026), checked 12 November 2026 ↩
Submission Guidelines, AI & LLM Policy (Annals of Mathematics), checked 12 November 2026 ↩
Artificial Intelligence: extracts from the Ethical Policy for Journals (London Mathematical Society, updated October 2026), checked 12 November 2026 a b
Generative AI policies for journals (Elsevier), checked 12 November 2026 ↩
AI use in manuscript preparation (Springer Nature editorial policies), checked 12 November 2026 a b
Using AI in peer review (Springer Nature editorial policies), checked 12 November 2026 ↩
Information for Authors (EMS Press), checked 12 November 2026 ↩