Editor’s Note: Today’s post is by Ch. Mahmood Anwar. Dr. Anwar is a research consultant, editor, author, entrepreneur, and HR and project manager. His research interests include critical analysis of published business research, research methods, new constructs development and validation, theory development, business statistics, business ethics, and technology for business.

The conversation around generative artificial intelligence (AI) in scholarly publishing has moved rapidly, from breathless enthusiasm to cautious policy-making. While publishers and editors have swiftly reached a consensus position that AI tools do not qualify for authorship because they cannot take responsibility, manage conflicts of interest, or hold copyright, a more insidious problem is emerging: the new ghost in the scholarly machine.

The key ethical threat is not the listing of “ChatGPT” as an author, which most major publishers expressly forbid. The true crisis lies in the undeclared, pervasive use of large language models (LLMs) to generate substantive portions of a manuscript, e.g., the literature review, the portions of methodology, the discussion, or the framing of a conclusion. When a researcher submits an article that is significantly drafted by an LLM without clear disclosure, they are effectively engaging in a contemporary form of ghost authorship.

Defining the New Ghost

Traditional ghost authorship refers to a substantial contribution to a publication by an uncredited, often paid, human individual. It fundamentally breaches the core principle that academic authorship must honestly and accurately reflect contributions and accountability. If editors identify ghost authorship in any manuscript, COPE suggests adding missing authors to the list. However, AI should not be credited as a co-author.

Concealing generative AI’s role in providing content to a scholarly publication is an example of this ethical breach, but with a technological twist. When an author relies on an LLM to generate the intellectual content, rather than just to polish the language, they are outsourcing intellectual effort without attribution. This creates critical gaps in accountability:

• The Responsibility Gap — An LLM cannot be held accountable for fabricated data, plagiarized text, or “hallucinated” citations that it generates. The human author is ultimately responsible for the integrity of the entire work. Undisclosed reliance shields the contribution from scrutiny.
• The Transparency Gap — Scholarly communication relies on transparency to build trust. When a reader, editor, or reviewer encounters AI-generated text disguised as human thought, the foundational trust in the research process is compromised. It masks the true origin of the ideas and the effort invested. 

The Problem of Pervasive Assistance

Most publisher policies distinguish between permissible and impermissible AI use. In most cases, AI is allowed for language and readability, but prohibited for generating, manipulating, or reporting original research data and results.

However, the middle ground — the narrative prose of the manuscript itself — is where these policies tend to break down. Is it merely “language improvement” to use an LLM to synthesize a literature review from ten uploaded papers, or to rewrite a complex results section to make it sound more authoritative?

This distinction, which determines whether disclosure is necessary, is notoriously fuzzy. A helpful framework (devised by the author), informed by the Committee on Publication Ethics (COPE) position, suggests a spectrum of AI use presented in the table below.

As seen in Category II, the ghost writer is not an uncredited human, but an uncredited algorithm that provided the structural framework and authoritative prose for the argument, leaving the human author as merely the final editor and submitter. Figure 1 provides the flow diagram for editors to check submitted manuscript for AI content and making desk decisions.

The Path Forward: Mandated Transparency and Policy Harmonization

To mitigate the threat of AI ghost authorship, the scholarly community needs to move past simple prohibition and toward mandated transparency, which the APA is now recommending for authors.

1. Standardized Disclosure Statements: Publishers must harmonize their requirements, perhaps using a CRediT-style taxonomy for LLM contributions (e.g., “Drafting – LLM-Assisted, 70% of Introduction”). This would provide the necessary granular detail beyond a simple “AI was used” type generic statements. For instance, Taylor and Francis request that authors include the full name of the tool used (with version number), how it was used, and the reason for use. Similarly, Elsevier urges authors to document their use of AI, including the name of the AI Tool used, the purpose of the use, and the extent of their oversight. However, the extent of AI assistance in idea conceptualization, drafting various section of manuscript, and the boundary conditions are rarely documented.
2. Citation of Prompts: When substantive content is generated, the human-authored prompt must be disclosed, either in the methods section or in a supplementary appendix. This allows reviewers and readers to understand the human’s intellectual input versus the machine’s output.
3. The “Humanity Test”: Humanity test is the term devised by the author to refer the accountability and verification requirement. Authors must be able to verify, reproduce, and take accountability for the content generated by the AI.If a section is so wholly machine-generated that the human author cannot defend its core claims or trace its sources, it fails the requirement for human accountability and must be rewritten or fully substantiated by the authors.

The challenge posed by generative AI is not just technical; it is a fundamental test of the publishing industry’s commitment to the principles of accountability, integrity, and transparency. If scholarly publishing allows the new “ghost in the machine” to persist through a lack of clear and enforceable disclosure, the trust underpinning the entire system will be irrevocably diminished. The tools are here; the ethics must catch up.