Authors Must Disclose AI. Why Don’t Journals Have To?
The new rule in academic publishing is simple, repeated, and morally loaded: if authors use AI, they must disclose it. Say where you used it. Say how you used it. Verify the output. Take responsibility.
That sounds fair — until you look at what is happening on the other side of the submission portal.
Springer Nature said on 12 March 2026 that in 2025, more than 1.5 million papers passed through workflows supported by nearly 60 AI tools, covering screening, editorial evaluation, retention, and research-integrity checks, with a further increase planned in 2026.
That means the modern author is entering a system where AI may help decide what gets flagged, routed, questioned, or escalated before a human editor has fully engaged with the manuscript. Yet the publishing conversation still treats AI disclosure mainly as an author duty, not as a shared transparency obligation across the whole system.
This is the contradiction no one should ignore: authors are being asked to explain their use of AI in detail, while journals and publishers are not equally required to explain how AI shaped the handling of a submission. That is not a minor policy gap. It is a fairness problem.
The rule sounds ethical because it is incomplete
Of course authors should not hide AI use. If a tool drafted sentences, rephrased arguments, suggested references, summarized literature, or generated images, readers and editors have a legitimate interest in knowing that. The recent JCOM editorial on generative AI and fabricated references makes the stakes plain: journals are seeing “ghost references” and warning that manuscripts containing fabricated content may be rejected, withdrawn, or retracted.
But notice how the burden is framed. The author must disclose. The author must verify. The author must be accountable.
Meanwhile, publishers are building AI deeper into the publishing pipeline and describing the result as “less friction.” That may be true from an operational point of view. But less friction for the system can mean more invisible scrutiny for the author.
This is where the language of responsibility becomes selective. When authors use AI, it is treated as a matter of ethics. When publishers use AI, it is often treated as a matter of workflow efficiency.
If AI helps judge a paper, shouldn’t authors know that?
This is the question I keep coming back to.
If AI materially influences triage, integrity escalation, reviewer selection, or editorial routing, why should that remain mostly invisible at the manuscript level? If a paper is pushed into additional checks because a system flagged something unusual, should the author not know that? If an integrity system incorrectly signals risk, what meaningful appeal path exists?
The usual answer is that there is still human oversight. But “human oversight” is becoming a comforting phrase that often hides the more important issue: how much human judgment happens before or after machine guidance has already narrowed the range of attention.
Anyone who has submitted to journals knows that timing, tone, and first impressions matter. If AI systems increasingly shape those first impressions, then authors deserve more than a vague promise that humans remain in the loop.
Reviewers are using AI too, often against guidance
Now the problem gets worse.
A Nature report said that a survey of 1,600 academics found that more than half had used AI tools while peer reviewing manuscripts, often despite journal guidance discouraging or limiting such use.
Pause there for a moment.
Authors are told to disclose AI use because trust matters. But many reviewers are already using AI in practice, and authors usually have no way of knowing when that happened, how much it shaped the report, or whether the reviewer followed journal rules at all.
That creates a serious asymmetry. The judged party is asked for transparency. The judging party often is not.
If a reviewer used AI to summarize a manuscript, improve wording, organize comments, or test the strength of an argument, should that be disclosed to the editor? To the author? If the answer is no, then the system is defending transparency as a principle only when it applies downward.
The trust problem is no longer theoretical
The JCOM editorial matters because it moves the debate out of theory and into editorial consequence. The editors warned that generative AI can produce plausible but false citations and reaffirmed that irresponsible use can lead to rejection, withdrawal, or retraction.
They are right to be concerned. In the AI era, polished prose is cheap. Scholarship is not.
That shift changes everything. For years, many editors and reviewers used fluency as one rough signal that a paper had been handled carefully. Now fluency can be manufactured at scale. A paper can sound intelligent and still contain invented references, weak synthesis, shallow reasoning, or borrowed confidence.
But here again, the solution being emphasized is mostly author discipline. Check every reference. Verify every claim. Declare your AI use. All sensible advice. Yet the system is simultaneously accelerating its own dependence on AI-mediated handling. The effect is that authors are being told to become more careful at precisely the moment the institutions around them are becoming more automated.
What exactly is still human in scholarly publishing?
An ACS Energy Letters editorial published days ago put the issue in a title sharp enough to stand on its own: “When AI Writes, Reviews, and Reads.” It is hard to imagine a clearer summary of where we are headed.
If AI can help write papers, help review papers, and help readers process papers, then the old boundaries of authorship and evaluation are dissolving faster than policy language admits. The editorial also notes that ACS Publications strictly disallows AI-generated reviews, which shows that even publishers who see the risk are struggling to hold the line in practice.
So what remains uniquely human?
Get Manuscriptedit Pvt. Ltd.’s stories in your inbox
Join Medium for free to get updates from this writer.
Is it authorship? Not fully, if drafting can be assisted.
Is it review? Not fully, if reviewers increasingly use AI anyway.
Is it editorial judgment? Not fully, if triage and integrity workflows are becoming machine-assisted.
Is it responsibility? Apparently yes — but mostly for the author.
That is the part that should make everyone uncomfortable.
This is not just a tech transition. It is a power transition.
The easiest mistake is to frame this as a tools debate. It is not.
This is a shift in who gets to see, sort, and shape scholarly work before it becomes visible to the world. AI inside publishing is not just speeding things up. It is redistributing power inside the evaluation chain. Publishers gain scale. Editorial teams gain operational support. Reviewers gain optional assistance. Authors gain new compliance burdens.
That is why the question “Should authors disclose AI?” is too small.
The harder question is: Why is disclosure being normalized as an author obligation, but not as a system-wide obligation?
If journals want to defend trust, they should be ready to answer a few uncomfortable questions:
· When did AI materially influence the handling of my manuscript?
· Did it shape triage, reviewer selection, or integrity escalation?
· What systems were used?
· What recourse do I have if they were wrong?
· And if reviewers used AI assistance, why am I not allowed to know?
Those questions are not anti-AI. They are pro-accountability.
What authors should demand now
Authors should absolutely verify their work more rigorously than ever. That part is non-negotiable. Check references manually. Keep records of how AI was used. Do not let polished language hide weak reasoning. Those are basic survival rules now.
But authors should also stop acting as if this is only a personal ethics issue.
It is also a governance issue.
If publishers can announce that AI now supports huge portions of manuscript handling, then they should move toward manuscript-level transparency standards of their own. If reviewers are using AI in reality, journals should say what disclosure they require and how they enforce it. If AI systems trigger suspicion, authors should have a visible route to contest those signals.
Because trust cannot survive as a one-way demand.
The real question
The real question is not whether authors should disclose AI.
They should.
The real question is whether journals, publishers, and peer reviewers are prepared to accept an equivalent standard of visibility about their own AI-mediated role in scholarly judgment.
Until the answer is yes, the current system will remain what it already looks like from the author’s side: a world where the person with the least institutional power is expected to carry the greatest burden of moral explanation.
Source note
· Springer Nature press release, 12 March 2026.
· Nature news report on AI use in peer review.
· JCOM editorial on generative AI, fabricated references, and ethical publishing.
· ACS Energy Letters editorial: “When AI Writes, Reviews, and Reads.”
