AI evidence extraction

How AI use in scholarly publishing threatens research integrity, lessens trust, and invites misinformation

Andrew Gray · Bulletin of the Atomic Scientists · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

3/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1080/00963402.2026.2628491

Methodology & findings

Study design

Narrative review and synthesis of existing literature and published reports.

Main result

The paper finds that "Some researchers estimate that in 2024, 13.5 percent of all papers in PubMed indexed journals had been processed using LLMs, representing around 200,000 articles that year" and that "researchers who have used LLMs in their writing produce around a third more preprints than their colleagues." Additionally, "identifiably LLM-edited papers are retracted twice as often as average." The study also reports that "28 percent of researchers said they had used LLMs for copyediting and 8 percent for generating new text, but half or more of both groups didn't disclose it in the paper."

Research paradigm

Critical interpretivism

Author conclusions

The authors conclude that "The scholarly publishing system is, undeniably, not in the best of health" and employ an immune system analogy: "The well-meaning use of AI to help speed things up might, in this analogy, be the fever that ends up sending the whole thing to its sickbed, opening the door for much more damaging illnesses-in the form of intentional and malicious disinformation-to take root and do real harm."

Risk of bias

Confirmation bias in selection of anecdotal examples; No systematic methodology for identifying papers discussed; Lacks quantitative assessment of the prevalence claims beyond cited statistics; No primary data collection to verify assertions; The paper is a narrative review without systematic methodology, vulnerable to selection bias in cited sources; Author's expertise in bibliometric tools may introduce perspective bias; The paper relies on cited research rather than primary data collection; Causal claims (e.g., AI use causing research integrity problems) are made from observational/correlational evidence; Selection bias in which papers/evidence are cited; Publication bias in reported statistics; Potential author bias given the author's stated expertise in bibliometric analysis of AI use; Limited systematic methodology for evidence synthesis

Open questions raised

The paper identifies uncertainty regarding: (1) whether intentional misinformation campaigns using AI-generated papers are already occurring; (2) how much AI has been used in any given paper since it is difficult to distinguish; (3) the extent of hidden biases in AI-assisted search systems; (4) how to reliably detect and quantify LLM involvement across different disciplines.
The paper identifies the need for better understanding of: (1) how to distinguish between LLM-generated and LLM-edited papers; (2) whether intentional misinformation campaigns using AI are already occurring; (3) how to detect biases or censorship in LLM-based search systems; (4) methods to restore resilience to the scholarly publishing system under AI-related stress.
The paper identifies the need for better understanding of: (1) the extent to which LLM-generated papers are being deliberately created for malicious misinformation campaigns; (2) the mechanisms by which AI-based search systems may introduce biases or unintentional censorship; (3) solutions for managing the increased burden on peer review and publication systems.

Extracted from: pdfAgreement 78%

Explore related topics

Epistemic Risk Research Integrity Conceptual