Your updated editorial policy requires authors to declare AI usage for higher-risk activities. How do you plan to enforce this disclosure in practice if peer reviewers and editors cannot reliably spot undisclosed AI involvement?

Declarations of AI use are required for all use, regardless of risk level. Disclosure needs to become a routine and expected part of the publishing process, and that’s what our policy and infrastructure updates are designed to support. Removing the stigma and making disclosure the norm rather than the exception is absolutely vital, and that’s something that I think is being recognised not only by us, but by stakeholders across the research ecosystem.

Safeguards, monitoring, and integrity checks remain important, and AI can itself be part of the solution by helping identify potential concerns at scale. But responsible AI use is ultimately about more than detection and control. It depends on clear expectations, transparency, accountability, and shared standards across the research community. Publishers have an important role to play in establishing such standards and applying them consistently. However, lasting solutions can only be achieved through collaboration among all stakeholders, including researchers, reviewers, editors, institutions, funders, and technology providers. Our policies will therefore continue to evolve alongside technological developments, regulatory requirements, and community expectations.

Is your publishing group offering specialized training for editors and reviewers on generative AI models to help them better identify structural hallucinations, synthetic patterns, or logical flaws?

We see supporting the research community in navigating AI as an important part of our role and take a holistic approach in doing so. Alongside our recent policy update, we provided practical guidance for authors, reviewers and editors. Additionally, we regularly engage with researchers, editors and other stakeholders across the scientific ecosystem, sharing expertise and offering training through forums, webinars, Q&A sessions and other activities that help individuals to navigate the complexities associated with the use of AI, and the assessment of said usage.

A core part of our reframing of the AI editorial policies is to broaden the lens through which we are viewing AI use, for research practice, manuscript preparation, peer review and manuscript assessment. Training and support for editors and reviewers should not solely focus on research integrity concerns and potential misuses of the technology. Indeed, training people to identify whether content was generated by AI is an impossible task and one with no meaningful end goal. AI-generated content cannot always be reliably recognised, and the ability to spot specific stylistic patterns is likely to become less meaningful as the technology evolves. The more important capability is rigorous critical assessment: evaluating whether methods are sound, evidence supports the conclusions, references are reliable, and the work meets scholarly and ethical standards.

Are you concerned that mandatory disclosure policies might unintentionally penalize honest authors due to reader or reviewer bias?

Transparency should not function as a warning label. The purpose of disclosure is to provide confidence through provenance and context, not to imply lower quality.

Our policy framework deliberately focuses on how AI was used, the impact it may have had, and the level of human oversight involved. Routine uses that support communication or efficiency are very different from uses that may influence interpretation, evaluation or decision-making.

We believe manuscripts should be assessed on their quality, validity and integrity, not on the mere presence of AI, or the assumption of AI use. By asking all authors and reviewers to declare any use of AI we are capturing important information which feeds into more informed, evidence-based evaluation and decision making.

The wider research community will need to continue developing shared norms and standards that encourage responsible disclosure without creating unnecessary stigma around appropriate AI use.

Does widespread AI adoption risk creating a more monotonous or standardized scientific literature?

Scientific literature is primarily about the quality of the research, the robustness of the findings, and the accuracy of their presentation, rather than individual writing style. In a pre-AI world researchers were in fact encouraged to adopt a specific tone and style in their writing. If AI helps researchers communicate their work more clearly and effectively, this is positive, with benefits including broadening participation, particularly for those writing in a language that is not their first, and making research papers more accessible.

Nevertheless, concerns about potential homogenisation are valid, especially in contexts and research fields where the crafting of the written argument is how the value and impact is delivered. This is precisely why it is important that AI augments rather than replaces human expertise. Research questions, hypotheses, interpretations, and conclusions must continue to be developed and owned by researchers themselves.

Ultimately, preserving diversity in scientific literature will depend on keeping human authorship, scholarly judgement, and accountability at the centre of the research process. As long as AI is used as a tool to support research and communication, rather than as a substitute for creativity and critical thinking, we do not need to consider its wider adoption as a negative path to a standardisation of the scientific record.

Will you be developing closed, enterprise-grade AI environments for internal editorial use and peer reviewers?

Confidentiality remains a fundamental requirement of editorial assessment and peer review. Unpublished manuscripts must only be handled in ways that protect authors’ work and maintain trust in the process.

We are continuously exploring how technology can help reduce unnecessary effort and support editorial workflows responsibly. All current and potential solutions need to meet our stringent requirements around privacy, security, transparency, and human oversight, and be assessed against our AI principles.

To give concrete examples: through Snapp, our in-house publishing and peer-review platform, we already provide a trusted environment that supports authors, editors, and reviewers throughout the publication process. This includes capabilities, some of which are AI-enabled, that help editors identify suitable peer reviewers, support the assessment of submissions against formal and editorial requirements, and assist in identifying potential integrity concerns such as problematic references, or other issues that may warrant further human review.

We are also developing AI-enabled solutions designed to help researchers navigate and engage with scientific literature and improve manuscripts for publication in a responsible and transparent way.

However, the key issue is not simply the environment in which AI is used. Even in a secure, closed environment, AI should support editorial processes rather than replace them. It is vital that we understand how the technology is used, what role it plays, and whether human judgement and accountability remain intact.

What fundamental boundary must AI never cross in scientific publishing?

For us, the fundamental boundary is accountability.

AI is likely to play an increasingly important role in supporting researchers with discovery, analysis, and communication. However, responsibility for the integrity of research, the validity of scientific conclusions, and decisions made throughout the creation and publication process must remain with people.

This is why our approach focuses less on whether AI was used and more on how it was used, what impact it had, and who remains accountable for the outcome.

What is equally important is that technology will continue to evolve, and policies will evolve accordingly. However, human judgement, transparency, and accountability must remain at the heart of research and scholarly publishing. AI can support these principles and augment human expertise, but it should not replace them.

AI disclosure statement: AI was used in the early drafting of these responses, but the substantive writing and editing were undertaken by the author, and full accountability remains with her.