Preliminary Recommendations from the Journals Working Group
Venues that disseminate scientific publications - such as journals, conferences, and preprint servers - should give authors choice and consent in whether and how LLM-based systems are used during the referee process. Prior to manuscript submission, such scientific venues should clearly communicate whether, and in precisely what way, LLM-based systems will be used during review, manuscript typesetting, and all other stages of the publication process. This commitment to transparency and the rights of authors applies equally to scientific venues themselves and to any third parties involved with the venue, such as editors, referees, and third-party copy-editing services. Authors should have the right to refuse this without being prejudiced in any way in the reviewing process.
Frequently Asked Questions
Q: What is the difference between choice and consent?
A: “Consent” is often used in a legalistic manner to indicate someone has signed something to indicate that they accept certain terms. But if your terms are “Consent to [whatever practice], or you have the ‘choice’ to not submit your paper to our journal” that is not really a choice at all. True choice means the authors can still participate in your process, publication venue, community, etc. even if they choose not to give consent.
Q: “But we are being flooded with submissions, we have to use LLMs to handle the deluge.”
A: Scientific venues are facing many pressures brought on by LLM use, including a tremendous increase in the number of submissions. Nevertheless, this does not relieve them of the obligation to be completely transparent with researchers about how submissions will be handled, and to treat authors with dignity and respect.
Q: What if the work is already publicly available (e.g. on a preprint server)?
A: Choice & consent should still be offered to authors without prejudice. Authors may wish to opt out of LLM use in reviewing for a variety of reasons, many of which may be completely orthogonal to the question of whether the work is already publicly available or even already explicitly used by an AI company in its training set. Scientific venues should not be forcing the response to such a choice on authors.
Q: What about using AI to detect plagiarism or to detect LLM-written papers?
A: Some AI tools may be useful for particular purposes, and we do not ask scientific venues to forego using them. (Although we note that “AI detectors” notoriously have many false positives, that disproportionately impact certain populations.) However, authors should be fully informed about the use of these tools before they submit their work.
Indeed, the right of authors to refuse need not be an all-or-nothing choice. Publication venues may wish to adopt a “menu” of possible choices of how LLM-based systems might be used or not used during the refereeing process, in order to better align author choices with reviewer and editor preferences. However, we emphasize again that care should be taken to ensure that such choices do not prejudice the outcome of the review decision itself.
Q: But we cannot enforce that our referees don't use LLMs.
A: There will always be rule-breakers, but this doesn’t make the rules less valid or important. Scientific venues are obliged to set expectations on referee conduct and enforce consequences when those expectations are violated.
LLMs are also so pervasive (e.g. in search “summaries”) that referees might accidentally use them without realizing it, even when trying to avoid them to adhere to the wishes of an author. We would recommend such accidental uses be ignored. As with many journal policies, a “best effort in good faith” criterion should be applied.
Q: What about papers that have already been submitted?
A: Publication venues that have neither been transparent with authors about the use of LLM-based systems in the reviewing process, and which may have allowed such use, should begin conversations with authors and referees as soon as possible. Such issues may, unfortunately, need to be addressed on a case-by-case basis.