Unpublished work and AI tools: what Nature's peer review stance may mean for surgeons

What was said

In a LinkedIn post on 28 September 2026, Magdalena Skipper, editor-in-chief of Nature, wrote that the post "is why we at Nature Portfolio ask our reviewers not to upload the papers they are reviewing for us to an AI tool". She linked to commentary on a New York Times report by Carl Zimmer about a recent claim from the AI company Anthropic.

Her statement is aimed at peer reviewers handling confidential manuscripts. Some commentators on social media read it as broader guidance for all scientists; others pointed out that it refers specifically to reviewers, who are handling work that is not their own. Readers may wish to read the original post in full.

The reported dispute

According to the post Skipper shared, Anthropic reported that its Claude agents identified a group of reverse transcriptases in jumbo phages whose function is not yet known. As reported by the New York Times, Mario Rodriguez Mestre of the University of Copenhagen said his group had studied the same systems for four years, and that he had shared unpublished findings, including a draft manuscript, with Claude while using it for coding and writing.

The same post reports that Anthropic said it knew of no published work describing these systems, that Claude was not trained on user transcripts, and that its biology team has no access to them. These are the company's own statements. It is not yet clear from public information whether the two pieces of work are connected, and ASTI has not assessed the claims.

Why it matters to publishing

Springer Nature's published editorial policies state that "confidentiality and data protection must be maintained when using AI tools" and that human accountability cannot be transferred to AI systems. Other publishers and editors have set out similar positions, noting that a manuscript sent for review is shared for confidential evaluation only.

There is also debate. Some academics have argued in preprints and online posts that reviewer bans are hard to enforce and that secure, enterprise AI tools may reduce the confidentiality concerns. This discussion appears to be ongoing.

Possible relevance for surgeons

Surgeons often hold unpublished material that may have value: trial data, device concepts, early prototypes, audit results and grant applications. Points that may be worth considering include:

• As a reviewer, the manuscript belongs to the authors and the journal; the journal's own AI policy applies.
• As an author or innovator, terms of service for AI tools differ, including on data retention and use; it may be worth checking these, and your institution's policy, before sharing unpublished work.
• Where intellectual property or patents may be involved, your employer's technology transfer or innovation team is the appropriate point of contact.
• Patient-identifiable information carries separate legal and governance requirements.

ASTI's forthcoming course on AI and academic publishing will cover some of these themes.

This article is an overview for information only. Readers should consult the original sources, journal policies and appropriate professional guidance.

Sources: this draft is based on reporting and statements published by the sources listed in the references below. Quotations and figures are attributed to those sources.