AI Pre-Screening Framework Proposed for Health Sciences Peer Review

A preprint on arXiv (2608.14625) introduces a novel framework for AI-assisted peer review in health sciences, titled 'Local AI pre-screening for human triple-blind peer review in health sciences.' This paper highlights the increasing pressure on academic peer review systems, referencing unprecedented submission rates at key conferences: NeurIPS 2025 with 21,575 submissions, ICLR 2025 at 11,603, and ICML 2025 with 12,107. The surge has surpassed the availability of qualified reviewers, resulting in undisclosed use of large language models (LLMs) in reviews. An independent review of ICLR 2026 indicated that about 21% of its 75,800 peer reviews were entirely AI-generated, an increase from 15.8% in 2024. Risks include fabricated citations and prompt-injection tactics to sway AI evaluations. The proposed system is a triple-blind, multi-LLM pre-screening process designed for a health sciences journal, ensuring transparency about AI use while keeping human reviewers as the ultimate decision-makers. It consists of five stages, starting with sanitization, to manage submission volumes while upholding integrity in peer review.