LLMs Threaten Double-Blind Review Anonymity
A recent study published on arXiv (ID 2608.05157) indicates that large language models (LLMs) possess the capability to de-anonymize authors of academic papers more effectively than humans, posing a threat to the integrity of the double-blind peer review process. The investigation, which analyzed only titles and abstracts from papers released after the model's training, demonstrated that LLMs narrow down potential authors to a limited group of five domain experts. This susceptibility remains even when stylistic and bibliographic elements are removed, suggesting that consistent patterns in how problems are framed and research topics are developed serve as hidden indicators of authorship. The results imply that the safeguards against bias related to status and affiliation in scientific publishing are becoming increasingly vulnerable due to LLMs.
Key facts
- Study posted on arXiv with ID 2608.05157
- LLMs can de-anonymize authors more efficiently than humans
- Used only titles and abstracts from papers published after model training
- Belief concentrates onto a small subset of plausible authors from pools of five domain expert candidates
- Vulnerability persists even when stylistic and bibliographic cues are excluded
- Stable patterns in problem framing and research focus function as latent conceptual signatures of authorship
- Double-blind peer review is the scientific community's primary defense against status and affiliation bias
- Findings indicate that double-blind review is increasingly fragile in the presence of LLMs
Entities
Institutions
- arXiv