AI and the illusion of a new crisis
Artificial intelligence has become one of the primary concerns in academic publishing, with headlines warning about AI-generated papers, fabricated references and the erosion of trust in the scholarly record. While these concerns are valid, they risk focusing attention on the wrong issue. The underlying challenges in academic publishing did not begin with AI, and they will not end with it.
The system was already straining a long time before these tools became widely available. Paper mills were producing manuscripts at scale, authorship was being bought and sold, citation manipulation was distorting metrics, and editorial systems were struggling to cope with increasing submission volumes. These issues were well established, even if they were not always openly discussed.
What AI has changed is not the existence of these problems, but the speed and scale at which they can be exploited. AI has reduced the cost of participating in behaviours that were already embedded within the system. Moreover, they are now even harder to control.
From production bottlenecks to production pipelines
Producing a research paper has traditionally required time, expertise and sustained effort. Even in cases where quality was questionable, the process of drafting, revising and formatting a manuscript created natural constraints on how quickly papers could be produced. These constraints limited scale, even in environments where incentives encouraged high output.
AI is slowly, but surely, removing many of these constraints. Draft manuscripts can now be generated rapidly, literature reviews can be assembled with minimal effort, and language barriers can be overcome almost instantly. Editing, formatting and even journal targeting can be supported through automated tools, creating a workflow that was not possible just a few years ago.
This does not mean that AI produces rigorous or reliable research. However, it does mean that the barrier to producing something that resembles a research paper has fallen significantly. In a system that relies heavily on signals such as structure, formatting and presentation, this distinction becomes important. If something looks like a paper and passes basic checks, it can enter the scholarly ecosystem regardless of its underlying quality.
The evolution of paper mills
Paper mills have long been described as industrial operations that produce manuscripts for researchers seeking career advancement. Their business model relies on demand created by evaluation systems that reward publication output, often without sufficient scrutiny of how that output is produced.
AI does not replace paper mills. Instead, it strengthens and extends their capabilities. With AI-assisted workflows, these operations can generate drafts more quickly, vary text to avoid detection and produce multiple versions of similar papers with minimal additional effort. The economics shift in their favour, as costs fall and output increases.
The tools that enable these practices are no longer confined to organised operations. Individual researchers now have access to technologies that allow them to operate, at scale, as a paper mill – either for themselves, or as a way to generate revenue. This creates a more decentralised environment, where the capacity to generate large volumes of low-quality research is no longer limited to dedicated organisations.
The result is a shift from centralised production towards a more distributed model, where the distinction between individual misuse and organised activity becomes increasingly blurred.
Synthetic literature and fabricated references
The issue of fabricated references has become one of the most visible concerns associated with AI in research. Examples of non-existent citations appearing in manuscripts have attracted significant attention, raising questions about the reliability of AI-assisted writing.
However, this issue should be understood within a broader context. Academic publishing has long placed value on the appearance of scholarship, where extensive referencing and structured literature reviews are seen as indicators of rigour. This creates an environment in which the inclusion of references is incentivised, sometimes regardless of their relevance or accuracy.
AI automates this behaviour by generating plausible-looking citations, citing them and putting them into the biography. While some of these references may be entirely fabricated, others may be real but poorly matched to the content. In both cases, the underlying issue is not simply the technology, but the limited capacity of the system to verify references at scale.
Reviewers rarely have the time to check every citation, and editorial processes are not designed to validate each reference in detail (although they should be). AI operates efficiently within these constraints, producing outputs that appear credible but may not withstand closer scrutiny.
Automated workflows and editorial pressure
The rapid growth in submission volumes has already placed considerable pressure on journals and reviewers. To cope with this increase, many publishers have introduced automated workflows to manage submissions, screen manuscripts and support editorial decision-making.
AI is now becoming part of this infrastructure. It is used to identify suitable reviewers, assess the structure of submissions and, in some cases, assist with the preparation of peer review reports. These tools can improve efficiency, but they also introduce new challenges.
If AI is used to generate manuscripts and also used to process them, the system risks becoming increasingly automated at both ends. Submissions can be produced quickly, reviewed under time constraints and processed through systems that prioritise efficiency. While each step may function effectively in isolation, the overall effect can be a reduction in meaningful scrutiny. Indeed, we could reach a point where papers are written by AI and reviewed by AI and the paper is never read by a human. This is worrying as these papers can then be used to train AI models, even though a human has never validated the contant.
AI as an economic force
To understand the broader impact of AI on academic publishing, it is helpful to view it through an economic lens. One of the most significant effects of AI is that it reduces the marginal cost of producing a paper. Tasks that once required substantial time and effort can now be completed far more quickly, lowering the overall cost of participation.
When production costs fall, supply tends to increase. In academic publishing, this means more papers are submitted, placing additional pressure on journals, reviewers and indexing systems. If demand for publications remains strong, driven by promotion criteria, funding requirements and institutional expectations, the system expands.
This combination of sustained demand and reduced production costs creates conditions in which output can grow faster than the system’s ability to evaluate it. The result is not just an increase in volume, but a growing challenge in maintaining quality and integrity.
The real question: what is being optimised?
Much of the current discussion around AI in publishing focuses on detection. There is considerable interest in identifying AI-generated text, detecting fabricated references and developing tools to screen submissions more effectively. While these efforts are important, they do not address the core issue.
The fundamental question is what the system is designed to reward. If researchers are evaluated based on publication volume, speed and presence in indexed journals, then tools that make it easier to produce papers will be used to meet those expectations. AI does not alter these incentives, but it enables them to be pursued more efficiently.
In this sense, AI is not driving behaviour. It is responding to it. The technology aligns with existing incentives, amplifying their effects rather than reshaping them.
Governance Lens: managing acceleration, not just technology
From a governance perspective, the challenge is not simply to manage AI as a new technology, but to understand how it interacts with existing systems and incentives. This requires a broader view that goes beyond technical solutions.
Institutions and publishers may need to consider how AI is being used in the production of research, what processes exist to verify references and data, and how peer review systems can adapt to increasing submission volumes. It is also important to examine whether current evaluation frameworks remain appropriate in an environment where output can be generated more easily and at greater scale.
These are questions of policy and oversight rather than technology alone. AI accelerates processes, but governance determines how that acceleration is managed and whether it supports or undermines the integrity of the system.
Where does this lead?
Framing AI as the central problem in academic publishing offers a clear and convenient narrative, but it risks overlooking the structural issues that shape behaviour. The reality is that AI amplifies both the strengths and weaknesses of the system in which it operates.
In a well-aligned system, AI can support researchers, improve efficiency and enhance access to knowledge. In a system where incentives prioritise volume over quality, it can scale behaviours that weaken trust in the scholarly record.
This leads to a more challenging conclusion. If the underlying incentives in academic publishing remain unchanged, the introduction of AI will not resolve existing problems. It will accelerate them, making it more difficult to maintain standards and ensure integrity.
Your thoughts
Artificial intelligence is now embedded within the research ecosystem, and its influence will continue to grow. The question is not whether it should be used, but how it should be governed.
If AI reduces the cost of producing papers, and incentives continue to reward publication volume and speed, what changes would be needed to ensure that quality, integrity and trust remain central to the system?
About the author
Graham Kendall is an experienced academic leader and researcher, with over 25 years in higher education and more than 300 peer-reviewed publications. He writes on research integrity, governance and the changing dynamics of scholarly publishing.
He works with universities, publishers and research leaders on issues including research evaluation, academic governance and publication strategy. If you are reviewing your institutional approach to AI in research or publishing practices, he welcomes the opportunity for dialogue.