Phantom references and the peer-review crisis: how artificial intelligence tests the resilience of scientific periodicals

Abstract

The critical vulnerability of the traditional blind peer-review system to the challenges posed by the rapid development of generative tools is examined. Based on a real-world precedent involving the discovery of a completely fabricated bibliography in a submitted manuscript, we analyse the basic mechanisms behind the creation of false references, such as anachronisms, “Frankensteinisation”, and professional biases. The study demonstrates the evolution of the threat: a transition from the obvious errors of early algorithms to the deep “semantic hallucinations” of modern RAG-based search engines, which are capable of generating perfectly formatted yet conceptually empty texts derived from real databases. To protect the publication process, an updated algorithm for editorial control is proposed, requiring the mandatory validation of digital object identifiers (DOIs) and a clear declaration of the algorithms utilised by the authors. The main conclusion emphasises the necessary and unalterable transition to the Open Science Framework paradigm, where textual material is viewed merely as an accompanying document to a verified array of primary datasets, open-source code, and deposited collection specimens.