Authorship represents far beyond the mere inclusion of a name on a manuscript; it signifies intellectual ownership of ideas and full accountability for the integrity and content of the research. Historically, the concept of authorship has been inseparably linked to human consciousness and originality.[1,2] However, with the rise of generative artificial intelligence (AI) in the form of large language models (LLMs), the boundary between human contribution and machine assistance has become blurred. Therefore, it is essential to fundamentally revisit the traditional concepts of authorship, accountability, and originality.[3] In this new landscape, the central question can be regarded as “What wrote this?” rather than “Who wrote this?”, as the answer most probably does involve an algorithmic assistant, particularly given their vital role in democratizing publishing and dismantling language barriers for non-native English speakers.[4,5] While we should make an effort to remove the stigma surrounding LLMs for manuscript drafting, a hard line must be drawn between passive oversight and true intellectual contribution. Supervising AI may create a false sense of competence; however, merely reviewing or approving AI-generated text falls fundamentally short of original authorship and cannot be regarded as an original intellectual contribution, since LLMs fundamentally recycle statistical patterns from existing literature, solely relying on them risks driving scientific discourse toward descriptive mediocrity. More concerningly, this dependence actively erodes the new generation of researchers' fundamental capacity to learn, conceptualize, synthesize, and think critically.
Academic credit remains grounded in the human-centered authorship criteria established by the International Committee of Medical Journal Editors (ICMJE) and Committee on Publication Ethics (COPE), where one principle is unequivocal: AI tools cannot be recognized as authors, as AI systems lack consciousness and legal personhood, they cannot assume ethical or legal responsibility for the content they generate, nor can they replace human expertise, particularly in tasks requiring innovation and critical decision-making.[2,6-8] As the COPE notes, AI is simply a “dancing bear” with no understanding of its own actions.[7] Until the stigma surrounding these tools dissipates, radical transparency must be the standard. Endorsing the ICMJE’s Revised January 2026 Recommendations, Joint Diseases and Related Surgery requires explicit disclosure of AI use during manuscript submission: in the Acknowledgments for drafting, the Methods for analysis, and the legends for generated images.[6,9]
Finally, the emerging phenomenon of “AI-giarism” also warrants attention.[10] Generative AI models inherently tend to plagiarize while simultaneously obscuring the origins of ideas, making accurate academic attribution and the governance of unintentional plagiarism increasingly challenging. Consequently, the heavy burden of verifying sources rests entirely on the researcher and policing AI output is incredibly inefficient. A critical question remains: Is it actually less work to police an LLM than to write the text yourself?
In conclusion, scholarly publishing is a collaborative endeavor that depends on scientific rigor, transparency, and integrity. Regardless of how sophisticated AI becomes, responsibility cannot be delegated to machines. Ultimately, the scientific question of “Why?” and the clinical question of “So what?” should remain unequivocally within the domain of human intelligence, judgment, and accountability.