Reproduce “"Give a Positive Review Only": An Early Investigation Into In-Paper Prompt Injection Attacks and Defenses for AI Reviewers” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Qin Zhou, Zhexin Zhang, Zhi Li, Limin Sun
The authors test whether hidden PDF instructions can raise AI peer-review ratings when text extraction preserves them. They compare fixed prompts with prompts refined using simulated reviewer feedback, and test detection. For DeepSeek-Chat on 100 sampled ICLR 2025 papers, the combined fixed prompt raised mean rating from 7.06 to 9.86, while iteration reached 10.00; these are model scores, not acceptance outcomes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.