For much of the past three years, variations of the same question have surfaced repeatedly in conversations with editors and peer reviewers: “I think the author has used AI. Should we reject the paper?”. Similar concerns have also been raised about peer review reports. The details differ, but the underlying assumption is often the same: that the presence of artificial intelligence is, in itself, the issue.

That assumption is understandable. When generative AI first emerged, publishers, universities and research organisations were faced with a genuinely new challenge. Existing policies had not been designed with large language models in mind, and there were (and remain) legitimate concerns about accuracy, authorship, accountability and even fakery. While we at Springer Nature – and many others – were clear that simplistic prohibition was an unrealistic and unhelpful approach, decisions about whether AI should be permitted, restricted or prohibited in particular activities inevitably came – creating a fairly overwhelming reluctance by the research community to disclose AI use.

Three years on, a taboo remains around disclosure, despite the fact that AI is increasingly embedded within the everyday tools researchers use to search the literature, analyse data, write code, draft and edit text and navigate information. We now understand how important it is to have clarity about how AI has been used, what impact it may have had, and whether appropriate human oversight and accountability have been maintained.

This challenge has been central to discussions surrounding Springer Nature’s new editorial policies and guidance on AI use. As we worked with editors, research integrity specialists and publishing colleagues to develop the framework, it became increasingly clear that governance based primarily on lists of permitted and prohibited technologies risks becoming outdated almost as quickly as it is written. It also became evident that unless users become more comfortable with declaring AI use and make consistent and accurate disclosures, it will not be possible to make confident judgements about submissions and the research activities to which they pertain.

ߣߣ��Ƶ

We therefore concluded that a more pragmatic approach is to focus on the principles that matter regardless of which tools happen to be available at the time: accountability, transparency, confidentiality, editorial independence and scholarly judgement. And it is clear that not all applications of AI deserve the same level of concern.

AI is already helping to make researchers more efficient and research more accessible, in ways that are entirely consistent with ethical and responsible use. There is a difference between using AI to improve readability, assist translation or help organise information, and using it to generate substantive interpretations, evaluations or decisions. One primarily affects efficiency. The other may influence judgement. The risks, and therefore the safeguards required, are different.

ߣߣ��Ƶ

From the moment we started developing our AI policies, we knew that authors and peer reviewers must remain accountable for the accuracy and integrity of their work – and editors for editorial decisions. Responsibility cannot be delegated to AI systems, regardless of how sophisticated those systems may become.

Transparency is equally important, and it’s an area where we see a lot of uncertainty and concern. Some researchers worry that declaring AI use may lead to the quality of their work being questioned, even when they have used AI in entirely appropriate ways. Others are unsure about where disclosure is expected or what level of detail they should provide.

If researchers continue to view declarations as grounds for suspicion rather than a normal part of scholarly communication, they will always be reluctant to make them – and AI use in meaningful parts of research design, practice or reporting will continue to go unscrutinised. For that reason, our new guidance places significant emphasis on declarations and creating greater clarity about how AI has been used.

This is not to normalise every application of AI, nor to imply that all uses carry equal risk. The aim is to support a wider research culture in which AI use can be discussed openly, assessed proportionately and evaluated according to its actual implications for research and research integrity, alongside encouraging all parties to take a more consistent approach. 

ߣߣ��Ƶ

Yet in a landscape shaped by rapidly evolving technologies, divergent perspectives and new regulations, the most important policy challenge is not deciding what is permitted but ensuring that everyone within the complex research ecosystem of funders, institutions, researchers, publishers and technology providers all speak the same language about what responsible use of AI actually means.

Expectations around AI use at the point of research design, funding application, institutional review and publication must be broadly coherent if researchers are to navigate them with confidence. Without that alignment, there is a risk that disclosure requirements are perceived as punitive, inconsistent or disconnected from the realities of research practice. Researchers are unlikely to be fully transparent about their use of AI if they are uncertain how that information will be interpreted or what consequences may follow.

Historically, many of the norms that underpin research integrity have emerged from the research community itself, with publishers, funders and others largely translating those expectations into their own policies. As publishers sit close to the end of the research life cycle, they have rarely been expected to lead systemic policy change.

AI has altered that dynamic. The pace of technological development, combined with the diversity of views on how AI should be used, assessed and disclosed, means that publishers’ policies must be continuously developed to reflect the concerns and challenges from the wider pool of stakeholders, while remaining grounded in ethical principles.

ߣߣ��Ƶ

Perfect consensus is unlikely to be a realistic goal. But we hope that by championing trust and transparency and maintaining our ethical principles and engagement with all stakeholders, we can ensure we have policy which reflects the nuance and challenge that AI has presented.

Ellie Gendle is head of journals policy at Springer Nature. AI was used to provide suggestions for an appropriate structure for a THE opinion article. The conceptualisation and the substantive writing and editing was undertaken by the author, and full accountability remains with her.

ߣߣ��Ƶ