Skip to content
Article Mohab AbdelKarim on Substack Sep 2026

Mohab A. Karim: how publishers are rewriting their AI policies in 2026

How publishers are rewriting their AI policies

Mohab A. Karim published this analysis on Substack on September 22, 2026. It covers how publishers — across journalism, academic journals, and book platforms — have revised their AI policies in response to the gap between written rules and actual author behavior. About 70% of journals now have some AI policy, yet only around 0.1% of papers disclose AI use, a gap the article treats as a compliance problem rather than evidence that AI isn’t being used.

The core finding: early policies were too vague to follow. The 2026 wave of rewrites moves toward specific permissions at the task level, rather than blanket approvals or prohibitions.

Permitted, conditional, and prohibited

Most publishers now permit AI with conditions rather than banning it outright. None accept AI as an author, on the grounds that a tool cannot take accountability for the work. The main dividing line is between assistive uses — refining your own writing — and generative uses — creating new content. Assistive use typically requires no disclosure; generative use typically does. Policies written under social pressure, such as the SFWA’s shift to a full ban, tend to be harsher and less precise.

AP standards as a reference point

The Associated Press updated its standards in July 2026. The AP permits AI for headline ideation, document summarization, transcription, translation, grammar review, and search optimization. Reporting, sourcing, editorial judgment, and verification remain human responsibilities. A journalist must review and edit every AI output before publication.

Where disclosure goes

One of the article’s most useful sections covers where each publisher wants disclosure placed. The answer varies enough that a disclosure correct for one publisher often ends up incomplete or misplaced after reformatting for another.

Elsevier requires a separate declaration section before the references. ICMJE recommends disclosure at submission — in the cover letter and in the work itself — with writing assistance in acknowledgments and data analysis in methods. Wiley requires disclosure at submission and in the manuscript at the relevant point. JAMA points to acknowledgments or methods depending on what the AI was used for. Amazon KDP requires disclosure of AI-generated content to Amazon rather than to readers, and counts translations as generated even when the underlying work is yours.

Stricter rules for images

Image rules are stricter than text rules across most of the publishers covered. Elsevier prohibits AI image generation outright. Elsevier, Wiley, and Springer Nature distinguish primary research images, data-derived visuals, and conceptual or illustrative material, with the strictest treatment reserved for primary research images. The reasoning is that a plausible but fabricated image can corrupt the evidence record in ways that readers — and sometimes reviewers — cannot easily catch.

Who it is useful for

Writers who submit work to academic journals, journalists at news organizations updating their AI standards, and editors responsible for drafting or revising AI policies. The article is well-sourced and specific enough to use as a reference when checking whether your own workflow is compliant with a given publisher’s requirements.