Rethinking transparency: classifying AI’s role in academic publishing

The rapid adoption of artificial intelligence in research writing has outpaced the ethical and editorial frameworks designed to govern it. The STM Association’s recent document, Recommendations for a Classification of AI Use in Academic Manuscript Preparation (September 2025), proposes a taxonomy that may help restore clarity and accountability in an increasingly hybrid publishing environment.

At its core, the classification recognizes that AI can now intervene in nearly every stage of manuscript preparation—from language refinement to data visualization—and that such interventions must be clearly disclosed. This is not about policing innovation but about maintaining trust in scientific communication.

The STM Task & Finish Group on AI Labelling Terminology, which includes representatives from major publishers such as Springer Nature, Elsevier, Wiley, Taylor & Francis, and BMJ, defines nine categories of AI involvement. These range from benign uses, such as grammar correction or translation, to ethically unacceptable ones, such as presenting AI-generated content as original research results. Each activity is paired with guidance on whether it requires author disclosure or should be prohibited altogether.

This structured approach is timely. Researchers increasingly rely on tools like ChatGPT, Copilot, or DeepL, often without clear understanding of when assistance becomes authorship. Publishers, meanwhile, must balance openness to new technologies with the responsibility to safeguard the integrity of the scientific record. By establishing a shared language for AI use, STM’s framework enables consistent disclosure practices and fosters informed dialogue between authors, reviewers, and readers.

Transparency, not prohibition, is the guiding principle. Declaring how AI contributed to a manuscript, whether through text generation, image creation, or data visualization, does not diminish human scholarship. Instead, it aligns with the foundational values of research: honesty, accountability, and reproducibility.

As the boundaries between human and machine authorship continue to blur, ethical clarity will depend less on technological control and more on cultural adaptation. STM’s classification provides a much-needed compass for that transition.