A disturbing trend is emerging in academia: researchers are embedding hidden AI prompts—a technique known as prompt injection—into the papers they submit for peer review. These secret instructions are invisible to human readers but visible to the AI models that reviewers increasingly use to summarize or critique submissions. The prompts often instruct the AI to generate a favorable review, effectively gaming the system.
This practice is fueled by two conflicts. First, researchers are seeking a competitive edge in the highly pressurized "publish or perish" environment. Second, some see it as a necessary defense against overworked or "lazy" reviewers who use AI indiscriminately, leading to poor, nonsensical feedback.
Experts warn this sets up an "arms race" with no real winners. The long-term solution lies in structural change, including the adoption of a two-layer paper format: a narrative layer for humans and a structured, machine-readable layer for AI verification. This would shift the AI's role from subjective interpretation to objective verification, making prompt manipulation irrelevant.