AI ‘Ghosts’ Plague Academic Publishing
google
| Source: Mastodon | Original article
A new preprint from Samsung and the University of Warsaw shows AI‑generated personas such as Elena Vasquez and Marcus Chen have been used as fake experts and co‑authors in hundreds of academic papers.
A new preprint from Samsung and the University of Warsaw uncovers a hidden layer of AI‑generated “ghost” authors that is quietly polluting the scholarly record. The study shows that large language models repeatedly invent the same fictional personas – for example Elena Vasquez and Marcus Chen – and insert them as volcano experts, astronauts, thriller protagonists, podcast hosts and even co‑authors in hundreds of independently produced documents. Because these names appear with legitimate‑looking DOIs on platforms such as ResearchGate and Zenodo, they surface in search results on Google Scholar and other aggregators, giving the impression of genuine scholarship.
The finding matters because it adds a new, non‑stylometric fingerprint to the growing problem of AI‑slop in academia. Earlier this month we reported on the broader chaos AI has caused in publishing, noting how fabricated references and low‑quality output are already overwhelming journal editors. The “ghost ensembles” identified by the Samsung‑Warsaw team provide a concrete way to trace the provenance of AI‑generated text, but they also reveal how easily fabricated authors can be propagated at scale, eroding trust in citation networks and inflating metrics that depend on author reputation.
What to watch next is how publishers, indexing services and research institutions respond. The authors suggest that monitoring recurring fictional name patterns could become a frontline detection tool, prompting platforms to flag or remove suspect records. At the same time, policy discussions are likely to intensify around attribution standards for AI‑assisted writing and the responsibility of LLM providers to curb systematic name generation. The next wave of research will test whether these detection methods can keep pace with ever‑more sophisticated generative models, and whether the academic community can restore confidence in the integrity of its published record.
Sources
Back to AIPULSEN