AI-assisted peer review may be vulnerable to abstract rewrites

A new arXiv paper warns that AI-assisted peer review may be vulnerable to low-cost manipulation. The authors find that superficially rephrasing a manuscript abstract can improve AI review outcomes without changing the underlying scientific content.

The result matters because AI tools are increasingly being explored for manuscript screening, reviewer assistance, and editorial triage. If simple wording changes can shift model judgments, peer-review systems could become easier to game.

The broader takeaway is that evaluation systems using LLMs need adversarial testing before they are trusted in high-stakes scientific workflows.