Accountability
Research ethics rests on identifiable people being answerable for claims. A tool cannot be answerable, which is why AI cannot be an author and why responsibility for AI-assisted text stays entirely with the authors.
The harder case is an automated editorial decision. If a manuscript is desk-rejected partly on an automated score, who is accountable, and what is the route to appeal? That question is largely unanswered in current publisher policy.
Transparency, in both directions
Authors are increasingly required to disclose AI use. Publishers disclose their own use far less consistently, and authors are often unaware that screening or scoring has been applied to their submission.
The asymmetry is difficult to justify on principle. If disclosure matters because readers should know how a text was produced, it matters equally for how a decision was made.
Equity
Language support is a genuine equity gain for non-native English authors. AI-text detection cuts the other way: detectors disproportionately flag writing by non-native speakers, so the same group benefits from the tool and is penalised by its policing.
Access is also unequal. Capable tools are increasingly paid, which risks a two-tier system where well-resourced groups produce more polished submissions.
Confidentiality
Unpublished manuscripts, patient-level data and material under review should not enter general-purpose tools. For peer review specifically, several funders and publishers prohibit it outright.
Where the line sits
A workable principle: AI may assist with expression, and the authors remain responsible for substance. Using a model to improve how a finding is stated is editing. Using it to supply the finding, the citation or the reasoning is not, and is where misconduct begins.
In short
What to take from this.
- AI may assist with expression; substance stays the authors’ responsibility.
- Detector-based accusations fall hardest on non-native English authors.
- Publisher use of AI deserves the same disclosure that authors are asked for.
Sources
- ICMJE and COPE positions on AI and authorship
- Funder prohibitions on AI use in peer review
- Evidence on detector bias against non-native English writing
Related