Introduction
To protect the integrity of military scholarship, the Scandinavian Journal of Military Studies (SJMS) is implementing a more restrictive AI policy (SJMS, 2026). In short, we require all creative and intellectual contributions to remain strictly human-driven. We are highly critical of authors outsourcing any part of the creative process or their intellectual contribution to generative AI (GenAI) tools, even in the smallest ways. We understand science as a human endeavour: bringing new evidence into conversation with existing research to advance our collective understanding. In what follows, we outline our reasons for issuing a restrictive AI policy, explain why we consider contemporary GenAI tools potentially harmful to research integrity, and discuss the ‘menial’ parts of the research process where we can accept some limited uses of this new technology in our journal.
Broadly speaking, the increasingly pervasive nature of GenAI tools poses a fundamental challenge to contemporary academic knowledge production. A “publish-or-perish” mentality, an artefact of the career pressures long characterizing the academic world, has often seen quantity tending to trump quality; the arrival of digital tools affording the possibility of accelerating the process of publication have made matters decidedly worse. Academic journals are widely experiencing a rise in submissions frequently overwhelming the peer-review process and the capacity of editorial teams (Andersen, 2026); the journal Organization Science’s recently published a study of submissions made since the release of ChatGPT in 2022, concluding that the rise in submissions – a staggering 42% – was achieved at the cost of a clear loss of quality (Gartenberg et al., 2026). And we are simultaneously facing the challenge of peer reviewers outsourcing their duties to GenAI tools as well (Williams, 2025). While we are lucky not to have experienced many instances of AI use by reviewers in our own journal, we are fully aware of the risk we are facing.
Overall, we see the threat of a vicious circle in academic publishing capable of eroding trust in what we do. Research produced with the help of AI tools is considered less trustworthy by the public, as readers struggle to discern the extent to which authors have used their own ideas or used AI tools to do the work for them (Chen & Jia, 2026). This jeopardizes the value of work conducted by human researchers, who see their output competing with machine-generated texts that pollute the academic ecosystem. Agentic AI bots can already use GenAI tools to produce quantitative social science papers – something likely to even further accelerate the output of researchers (Engzell & Wilmers, 2026). As a consequence, we assume that the incentives of academic career pressures will lead to an even greater rise in the quantity of scholarly articles submitted to journals such as ours in the not-too-distant future.
As the editors of SJMS, our capacity to mitigate the challenges to research integrity stemming from GenAI tools, let alone affect the broader trajectory in academic publishing, is limited. Nevertheless, we have a responsibility for what we can influence: the pages of our journal. We frequently experience issues with GenAI tools, ranging from submissions using “hallucinated” references to manuscripts in which significant parts of the creative process (for instance, survey or interview questions) were outsourced to GenAI tools. With this editorial, we aim to clarify that we consider such GenAI use to be inappropriate.
How GenAI tools work and why we reject their use in the creative stages of the research process
Before explaining our reasoning in more detail, we want to provide a brief discussion of how GenAI tools actually work.
“AI” refers to a set of technologies whose use can be legitimate or illegitimate, according to the purpose and the field of research (Wood et al., 2026). AI tools, for example, can be helpful for pattern recognition, something potentially very useful for STEM research. “Mechanistic” or “menial” research tasks like this will increasingly be performed with AI tools of all kinds. This does not have to be problematic; we are concerned, rather, with the advent of generative AI as a tool to which authors outsource substantial parts of the writing and research process. At SJMS, we place particular relevance on those creative aspects of research under particular strain from GenAI tools and their capability to create text. We want to publish military studies articles driven by human researchers’ knowledge, creativity, and ideas. And, therefore, we want to provide clearer guidelines on how contributors can navigate the growing significance of GenAI in academic publishing.
The design features of contemporary GenAI tools and the research practices they often inspire, we argue, are not suitable for the production of original research. And for this reason, they should be excluded from the creative process in military studies. Trained on vast, open-source datasets, GenAI systems like large language models (LLMs) simulate human language and written communication; rather than providing any original output, an LLM predicts a sentence’s most likely wording and structure according to its training data. It is, in this sense, a word-prediction machine, repurposing words from its training data into written responses that remain statistical calculations regardless of their accuracy or relevance.
Another important feature of most commercial LLMs is that they are fine-tuned and trained through so-called reinforcement learning, through which the preferences of users are baked into the model’s probabilistic word predictions. Put simply: an LLM is designed to provide the answer it has learned that the user is likely to want rather than the most objectively correct answer to a prompt. Hallucinations are thus not a malfunction; they are a natural consequence of how these systems are designed and trained. This leads to noted propensity for sycophancy, with LLMs seeking to reinforce rather than challenge the user’s biases and beliefs (Kwik, 2025).
The issues related to all of this are exacerbated by our collective tendency to “anthropomorphize” LLMs, perhaps because their output appears to be human. Doing so, we might mistakenly think the response to our prompt to be more genuine, original or insightful than it really is; we might attribute it an epistemological authority it does not deserve. Indeed, we know from decades of research on human-computer interaction that humans collectively suffer from “automation bias”: a willingness to delegate tasks to computational systems once a basic level of trust has been established, something which occurs in military contexts, too (Goddard et al., 2012; Horowitz & Kahn, 2024; Skitka et al., 1999). Importantly, automation bias might lead to academic deskilling, with users becoming less adapt at solving complex tasks without the aid of commercial AI systems; this can be expected to eventually erode critical and independent thinking – a crucial prerequisite for the production of original and impactful research (see, for example, Larson et al., 2024).
Taken together, we argue that the technological features of contemporary GenAI tools pave the way for unprecedented levels of research misconduct commonly manifested in fabrication, falsification, and plagiarism (All European Academies, 2023). GenAI tools are prone to generate the most persuasive evidence founded on non-existent data and sources. Exacerbating these issues further, the opacity of their processes affords users little insight into how and why the tools might have arrived at fallacious outputs. Finally, with often illegally obtained training data, LLMs plagiarise the work of human authors (The Associated Press, 2025).
For these reasons, we consider contemporary GenAI tools to be potentially hazardous to research integrity, and feel obliged to reject their use in the creative process – but, as we will set out in the next section, there are limited circumstances when tools of such utility may be responsibly used.
The Appropriate and Inappropriate Use of GenAI Tools For Research
GenAI is sold precisely on its ability to create content, and we want articles in SJMS to be founded on the creativity and intellectual quality of its human authors – but we are aware, nevertheless, that there are some grey areas of GenAI use. Here, we will clarify our position on the appropriate and inappropriate use of GenAI.
“If I have seen further,” Isaac Newton famously professed in 1675, “it is by standing on the shoulders of giants” (Turnbull, 1959, p. 416). Where new research draws on what came before it, we must insist on meaningful engagement. Any intellectual contribution made by its author, including the literature review, research design, and development of its methodological approach, should be a product of their own effort. We want the creative process, in other words, to remain entirely driven by humans standing on the shoulders of other humans.
This means that we do not want authors to use GenAI tools for tasks that require reflexive and problem-driven thinking. And this includes the use of AI tools to create literature reviews or for similar research assistance tasks.
To some, this might seem anachronistic. Yet we insist that authors who have not made the effort to read, comprehend, and summarize existing research themselves, instead outsourcing this crucial effort to GenAI tools, are not able to produce the high-quality contributions we seek for the journal. The same approach applies to the compilation of interview guides, qualitative coding of material acquired in interviews, analysis, and discussion. Again, we refer to research as a human endeavour. The interview guide should be driven by the research question, which, again, is linked to the literature review; the discussion should both refer back to the literature review and looking forward, to where knowledge is created. In short, we will not accept research articles in which authors have outsourced these parts of their research design to GenAI tools.
In our perspective, GenAI tools would effectively take the place of co-authors in cases where interview questions, survey questionnaires, or other crucial parts of the creative process have been generated by them. But when would GenAIs effectively take the place of authors? Defining authorship often happens in a grey zone, as prevalent definitions leave significant room for interpretation. The widely used Vancouver guidelines state that authorship should be based on four criteria:
1. Substantial contributions to the conception or design of the work; or the acquisition, analysis, or interpretation of data for the work; AND 2. Drafting the work or reviewing it critically for important intellectual content; AND 3. Final approval of the version to be published; AND 4. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. (International Committee of Medical Journal Editors, 2026)
The Vancouver guidelines thus theoretically allow abusive academic relationships in which senior academics let junior scholars produce the bulk of work but do not name them as co-authors because they have not included them in the final approval of the manuscript. In our view, this was highly problematic even before the arrival of GenAI tools.
While we are not able to shine a light on every equivocal problem of academic authorship, we nevertheless want to employ a modified version of parts of the Vancouver guidelines to show how we understand authorship – and the appropriate and inappropriate use of GenAI tools in this sense. In our (narrower) understanding of authorship, a person or a GenAI tool should essentially be considered a co-author if any of the two first criteria of the Vancouver guidelines are met. In other words, if they have made substantial contributions to the conception or design of the work; or the acquisition, analysis, or interpretation of data for the work; and/or they have drafted the work or reviewed it critically for important intellectual content. While no one would expect a GenAI tool to approve a manuscript before publication, these two criteria constitute a significant element of the writing process, for which human authors are responsible.
We consider it acceptable to use GenAI tools for improving your own ideas in limited ways, just as you would call on a colleague for informal feedback. In practice, this means that you can ask a GenAI tool to provide suggestions for improving weaknesses in an argument you have already developed yourself, but you cannot ask GenAI tools to develop an argument for you. It can also be appropriate to use GenAI tools for discovering empirical evidence or patterns in large databases – if the parameters for the search have been defined by the human author. In short, you can use GenAI in limited ways to develop and improve your ideas, but you cannot use it to make short-cuts in your creative process. GenAI must never be in the driver’s seat in any creative decisions you make.
SJMS is a multidisciplinary journal, open to publishing both qualitative and quantitative research. We have differentiated concerns about the use of GenAI across these two methodological approaches. In qualitative research, we share the view that contemporary GenAI tools carry considerable epistemic risks. These include the unreliability of the outputs they generate and the fallacy that their statistical calculations of likelihood are equivalent to systematic analysis (Jowsey et al., 2025; Nguyen & Welch, 2026). We are aware that software commonly used by qualitative researchers for coding empirical material, as well as transcription tools, may incorporate AI technology. We accept this, provided it is used only for pattern recognition in data produced through a human-led creative process. An example would be using a coding tool to highlight patterns in interview or survey data for which a human researcher has defined the parameters or categories, and who has developed the interview or survey questions without creative input from GenAI tools. The critical analysis of this raw data, however, must be conducted by a human. We therefore ask authors to declare what software they have used in their data analysis, so that we as editors can make an informed decision about whether its use was appropriate. If in doubt, please feel free to contact the editorial team before submitting your full manuscript.
We are also aware that quantitative research increasingly relies on software that incorporates GenAI-supported tools or add-ons that facilitate statistical analysis. We will continue to welcome quantitative research articles, and understand that somewhat tedious, or mechanical, aspects of the research process such as statistical coding invite the use of GenAI tools. While these tools can already perform such tasks well (Winberg et al., 2026) and will undoubtedly become better at them in the future, we remain concerned about researchers outsourcing the performance of statistical analysis to GenAI tools: authors need to be very well trained if they are to detect mistakes made by the systems they use (Schwarz, 2025).
While we accept that authors of quantitative articles may use GenAI tools for non-creative parts of their analysis, data analysis must remain transparent and reproducible, and those conducting the analysis must be skilled enough to be able to detect potential errors of the coding software. Moreover, the research endeavour must remain driven by humans. In other words, we do not want to end up publishing papers where authors have simply fed existing publications and datasets to GenAI tools or even agentic AI tools and tasked them with running a statistical analysis. We therefore require authors of quantitative research articles to carefully declare both their use of GenAI and the validation of the results upon submission. If we deem the use of GenAI tools in this process to be problematic, we reserve the right to reject the submission.
As an academic journal focused on military studies that offers a publication venue for practice-oriented articles, we are furthermore not blind to the reality of military organisations trying to assess the value of AI technologies, including GenAI, for operational purposes. We want to clarify that we still encourage submissions that approach different AI tools as an object of study. In our practice-oriented articles, this can mean that authors have to use AI tools for testing their possibilities and limitations in performing a certain task. For instance, authors may wish to assess the performance of an AI-powered battlefield management system or other software used in military operations. We do not consider this problematic, as long as AI tools are not the origin of any intellectual contribution and not used for designing the actual research questions or for other creative elements of the research process. Assessing the potentials and limitations of AI tools in the military profession already is a key concern for military studies (see, for example, Hunter & Bowen, 2024; King, 2025) and the importance of this research agenda will only increase in the foreseeable future. We will therefore continue to publish research articles on AI tools in the context of military operations and other military affairs (Vasankari & Koski, 2025).
It remains our belief, however, that military studies scholarship should maintain a critical distance to GenAI. It is becoming increasingly clear that AI tools used in planning or the conduct of the military profession have great influence on military professionals’ thinking about operational problems (Sookermany & Slensvik, 2026). As researchers, figuring out these issues and presenting possible solutions requires the kind of human judgment GenAI tools are not capable of by design.
With this new policy, then, we are not issuing a blanket ban on all AI technology in the journal. We will continue to accept work that uses GenAI tools in the very limited sense described above. We also allow GenAI tools to refine language or grammar, or for non-creative tasks such as interview transcription, while making authors aware that they will remain fully responsible for the content of their work. We want to allow authors with a non-native command of English to be able to use available tools to translate their own words.
Concerning language, we still strongly recommend authors not to feed entire papers into GenAI tools for reducing the word count, polishing the language, or for other editing tasks. GenAI has a certain style and tone, which we see in a large and growing percentage of our submissions. We would strongly prefer contributions to SJMS to be driven by the voices of our authors rather than the pallid language of GenAI tools. Being aware that declaring the use of AI for polishing language and similar tasks has been shown to inspire mistrust in readers (Chen & Jia, 2026), we insist that its use for language purposes is properly declared – but we appeal to authors to have faith in their own skills as writers rather than outsourcing editing tasks to GenAI tools that iron out their valuable idiosyncrasies in language and style.
Using AI for Peer Reviewing
The use of GenAI has also affected the integrity of the peer-review process. As problematic as existing review practices are, we as editors feel a certain responsibility to the scholarly community we serve in this respect as well. We do not want to contribute to an ever-accelerating cycle of artificially generated academic work that ends up being judged by artificially generated reviews. The output of GenAI tools can often appear plausible to non-experts, which places additional strain on peer review, since the work of assessment requires genuine experts. As imperfect as the current process of quality assurance in academic publishing is, it remains the only tool we have for approriately assessing the novelty of arguments, their relevance, and their scholarly accuracy. The increase in submissions noted above places this system under unsustainable strain. Colleagues in fields dominated by quantitative studies have experienced this pressure for longer than those in fields dominated by qualitative approaches. For many in the publishing industry, the logical next step is to find ways of using AI tools to scale up the capacity of the review process. While “an AI-edited journal publishing AI-generated manuscripts which have been reviewed by AI-algorithms does sound absurd” (Wiechert et al., 2024, p. 1463), it is not unreasonable to assume that this could become reality for a large proportion of academic research – and we do not want to be part of it.
Further, the discussion on GenAI and peer reviewing sheds light on another, often forgotten, aspect of the peer review process. Peer reviewing is not simply a question of assessing research. It is also a way of creating a solid research network, and an important part of community building. This benefit of the peer review process is particularly important in a multidisciplinary, if still relatively small, field such as military studies, consisting of many specialized researchers working across disciplinary boundaries, often with very few colleagues in precisely the same areas as themselves. With every review they write, the human peer reviewers in SJMS strengthen the nodes of a complex research network. Introducing AI agents into peer review would deprive us of this important benefit of the assessment process.
We already have a restrictive AI policy for peer review, which explicitly prohibits uploading manuscripts to any GenAI tool. Submitted manuscripts are treated as confidential, and for that reason uploading a manuscript for review to a GenAI tool is not permitted under any circumstances, under both Ubiquity and SJMS guidelines. This policy is intended not only to protect our authors’ intellectual property but also to ensure they receive adequate feedback from experts rather than formulaic reviews produced by probabilistic text generation. Our authors deserve to be evaluated by experts in the field. Moreover, we encourage our peer reviewers to alert us if they notice traits characteristic of GenAI-generated manuscripts – a lack of analytical coherence, for example; overly general summaries of existing scholarship – and, where appropriate, to press authors for clarification (Roberson, 2026).
From Policy to Practice: Shaping the Future of SJMS Together
We are aware that our AI policy is hard to enforce and that we have few tools available to guarantee that our contributors follow our rules. Since we cannot reliably test submissions for GenAI input, we will have to trust our contributors and peer-reviewers to follow our guidance – just like other journals who are opting for restrictive AI policies (Hedžet & Leese, 2025). So we appeal to our community to uphold the principles of research integrity amid the pervasive use of AI. Contemporary GenAI tools are, for the moment at least, not conducive to the production of the original, high-quality, research findings we seek. Indeed, the unconsidered and indiscriminate use of GenAI may even weaken creative and critical thinking among researchers of military studies.
This is not to say that, in the longer term, no value is ever to be gained from GenAI use in research. We base our AI policy on the current state of development of the technology. For that reason, we will remain both vigilant and curious about how to foster ethical and responsible forms of use. This also means that, while we currently believe a restrictive GenAI policy is necessary to ensure the scientific integrity of research published in SJMS, that position might change in the future. And if it does, we will revise our policy.