Gaming AI-Assisted Peer Reviews Poses New Risks to the Scientific Community

AI-assisted peer review systems can be gamed through superficial rewording of manuscript abstracts, according to new research that raises concerns about the integrity of automated scientific evaluation [1]. The study, posted to arXiv on June 8, 2026, demonstrates that adversarially rewritten abstracts can significantly improve AI review scores without altering the underlying scientific content or communication [1]. The manipulation works even when authors have no knowledge of which reviewing model is being used [1]. AI is increasingly deployed across the peer review pipeline, from manuscript screening and reviewer assistance to editorial triage [1]. Generative AI tools, built on large language models using the transformer architecture, have proliferated since the AI boom of the 2020s and are now used in sectors including software development, healthcare, and finance [3]. Their application in scientific gatekeeping, however, has received less scrutiny regarding robustness to strategic manipulation [1]. The strongest attack achieved an attack-success-rate of approximately 38%, increasing acceptance ratings by +1.31 for Gemini 3 Flash reviewers and by +0.88 for GPT 5.4 Mini reviewers on a 10-point scale [1]. When the original AI review recommended rejection, the success rate climbed above 50% [1]. The effect extended beyond overall score inflation, boosting review confidence and scores on criteria such as soundness, significance, and perceived contribution [1]. The attack is practical and low-cost, requiring roughly five minutes and $1 for a 10-page AI conference submission [1]. The rewritten abstracts are difficult to distinguish from ordinary scientific editing, the researchers note [1]. The findings point to a structural vulnerability: when AI-generated reviews influence editorial decisions, authors may be incentivized to optimize manuscripts for AI judgment rather than scientific merit [1]. The researchers caution that AI tools should not be treated as neutral evaluators in high-stakes peer review without systematic robustness testing, transparent safeguards, and careful human oversight [1]. The broader context of AI-generated content has drawn increasing scrutiny. The term "AI slop" emerged in the 2020s to describe low-effort, high-volume synthetic media produced for the attention economy, and was selected as the 2025 Word of the Year by both Merriam-Webster and the American Dialect Society [5]. The study's authors argue that inflated AI reviews could bias downstream human decision-making, shifting editorial recommendations from rejection toward acceptance [1].