Alignment Journal

Editorial policy

AI usage

Last updated: 17 October 2026

Note: Peer review has been greatly disrupted by the rise of large language models (LLMs). This policy represents a preliminary approach by the journal to balancing the benefits and drawbacks. We expect to revise it in the coming months in response to experience and technological developments, and you can help us: If the main body of your manuscript contains substantive AI prose, but in your judgment meets all other criteria, you can still upload it and show us what we’re missing.

AI usage policy for submissions

AI usage at the journal is governed by these principles:

  1. Authors are welcome to use AI tools such as LLMs to perform their research.
  2. AI use must be disclosed when it contributes substantively to the underlying research, and such disclosure should describe the usage with enough detail to make the research process transparent.
  3. The actual text in the main manuscript body should be written by the authors themselves subject to the following:
    1. AI tools are permitted for light and relatively sporadic editorial refinement of author-written text (spellcheck and some wording suggestions are fine, sentence and paragraph-level rewrites are not).
    2. AI generation is permitted for non-text: tables, plots, graphics, etc.
    3. Appendices and Supplementary Material may be AI written. This must be disclosed in the manuscript.
    4. The main text must be author-written and contain the complete motivation, main results, and a high-level overview of how the results were obtained. The importance and insight of the paper will be judged by the contributions presented in the main text. The correctness will be judged based on the paper as a whole.
  4. Authors are responsible for the entire contents of the manuscript and must ensure that they are correct, original, and readable with normal effort, regardless of whether a tool was used or disclosed.

Substantive AI usage

Substantive AI usage requiring disclosure includes:

  • generating factual claims, citations, or other content whose correctness requires verification;
  • drafting or organizing prose beyond local changes, such as determining the structure or argument beyond a few consecutive sentences;
  • producing code whose correctness could affect reported results or conclusions;
  • performing nontrivial mathematical manipulations or derivations;
  • generating figures or tables; and
  • contributing materially to the methodology, analysis, interpretation, or conclusions.

AI contributions to the following need not be disclosed since it would not be considered substantive:

  • spelling, grammar, or local wording suggestions;
  • formatting or purely cosmetic code;
  • routine checking of calculations and trivial mathematical improvements; or
  • use analogous to a conventional web search—such as locating terminology or sources—provided the authors independently evaluate and verify the resulting information.

AI usage policy for the review process

The policy for authors, reviewers, and editors during the review process is analogous:

  1. Using AI tools during the review process is welcome
  2. The actual written contributions to review must be written by the participant
  3. AI usage must be disclosed
  4. Everyone is responsible for the correctness and quality of their contributions

Once a narrow class of AI usage is disclosed by a participant (e.g., checking for mathematical errors), it is assumed to apply to their subsequent comments; participants need not repeatedly disclose similar usage in every subsequent comment.

Editorial verdicts (accept, reject, or revise & resubmit) appear as special comments in the review discussion. To the extent AI was used for that verdict in a way that went beyond any previously disclosed usage, the editor is responsible for disclosing AI usage there in the verdict comment.

Even if reviewers use AI for low-level assistance when reading a manuscript (e.g., intelligent search of the document, defining terms, etc), we strongly encourage them to form their own initial opinion of the high-level value of the manuscript before seeking AI input on it.