Original title: Governing AI Research Through Peer Review: A Mixed-Methods Study of the Longitudinal Effects of Ethics Flags Across Resubmissions
Kento Nishi, Alec Laprevotte, Isaiah Bullock, Mfoniso Andrew · September 2026
Big artificial intelligence conferences now use ethics flags during peer review to try to guide scientists toward safer, more responsible practices. Researchers wanted to know if this peer review process actually changes the research. To find out, they tracked rejected or withdrawn papers that received ethics flags, checking what authors changed when they republished them later. They discovered that scientists usually tweak how they write about their projects rather than altering their underlying research methods. Authors frequently agree with reviewers during the review debate to help their papers get accepted. Yet once a paper is rejected, they often discard those concessions if they do not consider the critiques valid reasons to fix the underlying work. Instead of shifting their actual research directions, researchers treat review as an editorial step that controls which stories get seen, changing how papers are presented without altering what they actually build or study.
A home cook enters a baking contest and a judge warns that the cake is too dense. The cook nods, promises to fix it, and while the judge is still at the table, the matter seems settled. Once the judge leaves, the cook enters the same dense cake in the next contest, only with lighter frosting and a new name. Here, AI researchers often respond to ethics concerns during the debate with reviewers, but after rejection they resubmit the same methods in a rewritten paper.
AI systems affect real people, and conferences are a major gate for what gets built. If ethics review only changes how papers are described, unsafe research can resurface unnoticed at the next venue. Showing past ethics flags at resubmission would let reviewers hold authors accountable.
The abstract doesn't say what happens to research before it is submitted, or to papers accepted despite ethics flags. It tracks only rejected and withdrawn ICLR submissions that later resurfaced publicly, so projects that disappeared are not covered. The author interviews involve 25 cases and are qualitative, so they don't prove how common each motivation is.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Governing AI Research Through Peer Review: A Mixed-Methods Study of the Longitudinal Effects of Ethics Flags Across Resubmissions — Kento Nishi, Alec Laprevotte, Isaiah Bullock et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.