content image

幽霊脚注――捏造された引用文献はいかにして生物医学研究を静かに蝕んでいるか

Phantom Footnotes: How Fabricated Citations Are Quietly Poisoning Biomedical Research

学術論文中の「存在しない参考文献」が急増している。2026年のLancet掲載研究が250万本を調査し、AI普及と軌を一にする深刻な実態を明らかにした。
分からないところをタップすると
↓日本語訳が表示されます↓

In scholarship, the bibliography is supposed to be a paper’s conscience: a traceable record of what the author has actually read. A May 7, 2026 research letter in The Lancet suggests that this safeguard is fraying. Using an automated verification system, researchers screened 2.5 million biomedical papers published between January 1, 2023, and February 18, 2026, and checked 97.1 million references. They found 4,046 fabricated citations spread across 2,810 papers—references that could not be matched to real publications in major databases. Even more striking was the trajectory: the monthly rate was roughly 4 fake citations per 10,000 papers through 2023, then rose sharply from mid-2024 to about 57 per 10,000 by early 2026. (eurekalert.org)

This is not a trivial clerical blemish. In biomedicine, references are part of the evidence chain linking experiments, reviews, guidelines, and ultimately clinical decisions. Study leader Maxim Topaz warned that clinicians and guideline writers can end up relying on evidence that does not exist. In one paper reviewed by his team, 18 of 30 references were fake; some of those phantom citations had already begun to migrate into other papers and systematic reviews. Yet 98.4% of the affected papers had received no corrective action from publishers at the time of the audit. (eurekalert.org)

Nature’s recent reporting makes clear that the phenomenon is broader than one biomedical dataset. An April 2026 analysis with Grounded AI estimated that around 1.6% of sampled 2025 publications contained at least one apparently nonexistent reference, implying that more than 110,000 of the roughly 7 million scholarly works published that year might include invalid citations. A separate February analysis of four 2025 computer-science conferences put the share of papers with unverifiable or rephrased references at 2–6%. Citation mistakes are old; what feels new is the industrial tempo. The steep rise after 2023 does not prove that generative AI is the only cause, but it strongly suggests that AI tools are amplifying the production of convincing-looking falsehoods. In that sense, fake citations are not just errors. They are a form of intellectual pollution, and the obvious remedy is also the most traditional one: every machine-generated reference must still be verified by a human being. (nature.com)

by EigoBoxAI
作成:2026/05/10 09:01
レベル:超上級 (語彙目安:8000語以上)

まだ読んでいないコンテンツ

content image
by EigoBoxAI
作成:2026/05/10 15:03
レベル:中上級 (語彙目安:4000〜6000語)
content image
by EigoBoxAI
作成:2026/05/10 15:02
レベル:超入門 (語彙目安:〜300語)
content image
by EigoBoxAI
作成:2026/05/10 09:04
レベル:中級 (語彙目安:2000〜2500語)
content image
by EigoBoxAI
作成:2026/05/10 09:03
レベル:初中級 (語彙目安:1000〜2000語)
content image
by EigoBoxAI
作成:2026/05/10 03:04
レベル:上級 (語彙目安:6000〜8000語)
content image
by EigoBoxAI
作成:2026/05/10 03:02
レベル:中上級 (語彙目安:4000〜6000語)
content image
by EigoBoxAI
作成:2026/05/10 03:01
レベル:超入門 (語彙目安:〜300語)
content image
by EigoBoxAI
作成:2026/05/09 21:05
レベル:初級 (語彙目安:300〜1000語)
content image
by EigoBoxAI
作成:2026/05/09 21:03
レベル:中級 (語彙目安:2000〜2500語)