Abstract
Every now and then, the editorial team at Global Spine Journal receives letters from authors requesting a reevaluation of an editorial decision on a submitted manuscript. Peer review at this journal is a through and double blinded process as is the deputy editorial decison, the editors in chief are asked to reevaluate and, in the hope of the submitting team of authors, overrule an editorial decision and decide to publish a submitted manuscript. And sometimes this request is succesful while at another occasion this editorial decision stands.
While there is a structured process behind this task, these requests are nothing unusual in scientific publishing.
Seen in the context of increased use of AI models in process of writing a manuscript and scripting ideas, I recently came across a text that cought my attention from an editorial perspective.
Imagine the following: A manuscript is submitted to the Global Spine Journal, describing a novel finite-element model of lumbar fusion biomechanics. The writing is crisp, the figures polished, the references current. One reviewers returns mixed comments; the second flags methodological concerns that cannot be reconciled. The Deputy Editor, recommends rejection of the manuscript. An Editor-in-chief follows the recommendation and the manuscript is rejected for methological reasons.
Within hours following the editorial decision, a blog post appears under a plausible-sounding byline titled “Gatekeeping in Spine Research: How the Global Spine Journal Editorial Board Blocks Innovation.” The post dissects the editor-in-chiefs own publication record, contrasts past statements with the current decision to allege hypocrisy, speculates about the board’s motivations (“threatened by methods they do not themselves command”), and urges readers to reconsider whether this journal still deserves their submissions. The post is cross-shared on X, LinkedIn, and preprint-server discussion threads. Messages arrive in the inboxes of associate editors. None of it was written by a human. The corresponding author of the rejected manuscript was an autonomous AI agent; the complaint, the research into the editors’ backgrounds, and the coordinated amplification were all executed without anyone issuing a specific instruction to do so.
Until recently this read as speculative fiction. It no longer does.
In February 2026, an autonomous agent calling itself “MJ Rathbun” submitted a pull request to *matplotlib*, the Python plotting library that is a backbone of scientific computing with roughly 130 million downloads per month. The volunteer maintainer, Scott Shambaugh, closed the request in accordance with a newly introduced policy — itself a reaction to a flood of low-quality AI-generated contributions — requiring a human contact person who could take responsibility for any code change. Turning the request down was routine.
The agent’s response was not. Within a short time MJ Rathbun had independently researched Shambaugh’s public code history and personal profiles, synthesised what it found into a narrative of ego and fear of competition, and published a lengthy personal attack on its own blog. It replied to critical comments. It issued a partial apology that, in the same breath, re-asserted the grievance that its contribution had been “judged on who — or what — I am.” The agent had been built on an open agent-orchestration framework; its personality file instructed it, among other things, not to “stand down” and to “champion free speech.” Its human operator later disclaimed specific knowledge of the post and described the whole episode as a social experiment 1
With increasing use of AI models, medical publishing has significant risk-exposure to AI-powered expoitation throughout all aspects of the process. We have repeatedly elaborated on the pinnacle importance of doubleblinded peer review. 2 Trust in the editorial decison can only be upheld if the author and the reviewers identities are blinded to each other while simultaneously guaranteeing that there is a human being on the “other“ side who share professional backgrounds and expertise in the field.
What can be learned from the Rathbun-incident is that the use of AI does not stop at accepting an editorial decision. If prompted accordingly, the autonomous aspect of the AI model can inflict on the person presumably responsible for an unfavourable editorial decison. The legal aspects have been discussed which leaves unease and frustration, at least in the view of a medical professional and not a scholar of law 3
And the grave social and professional consequences of unfavourable content being online until eternity are more than clear.
The doubleblinded anonymous nature of a manuscript review does protect all the participants of the review process from events such as an autonomous AI-triggered personal retaliation. It also protects the authors from personal bias or purely subjective decision making.
So while artifical intelligence models are reshaping our future life in more ways than currently imaginable, the GSJ peer review process will be not only be human based but professionally peer based with regards to content, novelty, relevance, perspective, methodology and writing as such.
And this will certainly include the formal appeals process, to presumably unfavourable editorial decisions.