IEEE Access Flagged Two AI-Written Papers: 12 Fake References in One, Preprints in the Other

IEEE Access flagged two AI-written papers submitted by author Benjamin Skuse in a test published by IEEE Spectrum on October 27. Neither paper was accepted or sent to peer review.

Skuse wrote in IEEE Spectrum that he submitted both papers to the journal under his own name. The first used the free versions of ChatGPT and Gemini on optical solitons, a field he studied for his Ph.D. The second used FARS, the Fully Automated Research System built by Singapore-based Analemma, with the AI disclosure removed.

How each paper was caught

The first paper failed on references. IEEE Access's integrity checks, which include the STM Integrity Hub and a MolecularConnections checker, flagged 12 references as fabricated or unverifiable. Another checker could not validate 24.

Jenny Mahoney, managing editor of IEEE Access, said: "Your fake article had lots of hallucinated references". She said that "so many hallucinated references immediately cast doubt on the entire work".

The second paper passed that test, because FARS cited real sources. IEEE flagged its heavy reliance on non-peer-reviewed preprints. Mahoney said that "our team would have rejected this paper because it did have a huge percentage of references to preprints". She also pointed to a figure label reading "lhewl" and stray dots in an AI-generated brain image.

Paper	Tool	Why IEEE flagged it
Round 1	Free ChatGPT and Gemini	12 flagged references, 24 more unvalidated
Round 2	Analemma's FARS	Heavy use of preprints, a nonsense figure label

What FARS can produce

FARS has a public record of its own. Analemma-led authors posted a paper on arXiv dated July 1, 2026, titled "FARS: A Fully Automated Research System Deployed at Scale". Its first public deployment produced 166 complete research papers across 67 AI and machine learning topics.

The authors evaluated the corpus with 282 structured reviews from volunteer reviewers covering 140 papers. They wrote that the system can produce "review-worthy and occasionally strong AI/ML research artifacts". The same reviews exposed "recurring failure modes in narrow experimental scope, methodological limitations, and integrity issues".

Skuse's article reports that FARS produced a research proposal in 21 minutes and a paper in about 2 hours. It also gives a figure of 166 papers in 417 hours. The arXiv abstract gives the 166 but not the hours, so the hours figure rests on Spectrum alone.

The volume problem

Skuse cites 57% of published academic articles in 2025 as showing detectable LLM influence, up from 12% in 2023. He says roughly 1 in 40 papers with statistical evidence of AI use formally disclosed it. IEEE Access receives about 60,000 submissions a year and accepts about 20%, around 12,000 papers.

Amanda Sulicz, research-integrity manager at IEEE, said: "AI slop is completely eroding trust in science and every scientific process that we've worked for centuries to create."

Xiwen Su, manager of product operations and partnerships at Analemma, framed it differently: "FARS doesn't kill the paper, but it does challenge the paper as the unit of knowledge". Ivan Oransky, cofounder of Retraction Watch, called AI "the accelerant" in publishing problems.

Why the catches matter

The two papers were stopped by different checks. The first tripped on fabricated references. The second used real sources, and the flags were its reliance on preprints and its figures. Review overhead growing with machine output is examined in The Terminal's piece on Brooks's law and AI agents.

Skuse's article says the average number of reviewer invitations per article has risen 50% since 2018. Checking machine output is also the subject of Surge AI's Dayjob: Tax benchmark.

Skuse's article does not say whether IEEE Access will change its screening, and it gives no date for any change.