In the last five or so years, large language models (LLMs) have transformed from a novel regurgitator of haphazardly stitched together sentences to an almost \u2018human\u2019 personality standing by our side as we tackle life. Whilst the perceived humanity of these models is the topic for perhaps a future blogpost, it is almost undeniable to understate the impact of LLMs in our daily lives. Do you need someone to proofread your essay you\u2019ve spent hours drafting? GPT (or one of its many counterparts) has you covered. Need help drafting an email from scratch? No problem. Want to write and\/or heavily edit an entire academic article which would typically require days, if not weeks, of research? Surely just needs a push of a button\u2026 right?<\/p>\n\n\n\n
Despite tremendous advances in LLMs, key issues mean they are not yet a fully dependable addition to our writing endeavours. They are known to fail when asked to generate new <\/em>content with only a basic prompt. Some of these failures have made headlines 1<\/sup>. Some of the scariest instances are those of hallucinated information <\/em>2\u20134 <\/sup>. This refers to the phenomenon where AI tools generate convincing information which is factually inaccurate or simply fabricated 2 <\/sup>. In Belgium, the Ghent university rector came under fire for citing quotes, supposedly from influential thinkers, which were later found to be AI-hallucinations 1<\/sup>.<\/p>\n\n\n\n Whilst there are numerous examples of the poorly cited and often AI-hallucinated papers falling through the cracks of the peer-review process, today we focus on a Frontiers in Microbiology <\/em>reviewtitled \u2018Broadly neutralizing monoclonal antibodies against influenza A viruses: current insights and future directions\u2019 <\/strong>5<\/sup>. This paper attempts to provide an overview of the current landscape of monoclonal antibodies (mAbs) which are being developed to confer protection against influenza A, highlighting \u2018technological advances, clinical performance, and scalability\u2019. However, it contains many of the hallmarks of text that has been created or edited with generative AI, despite the generative AI statement stating \u2018The author(s) declared that Generative AI was not used in the creation of this manuscript.\u2019<\/em><\/p>\n\n\n\n\n\n\n\n There were some recurring themes which seemed rather odd at first glance. For instance, there was a continued use of the term \u2018anti influenza virus\u2019. Although the thought of genetically modified viruses being used in the war against influenza is exciting, it\u2019s more science fiction than rigorous science (at least for now?). A likely explanation is that the authors meant \u2018anti influenza antibodies<\/strong>\u2019. Whilst this can be a simple oversight, or even a slightly inattentive use of the Replace All <\/em>command, it also points to a failure mode in LLMs, where the next token prediction is based strongly on the previous word. In a fairly large number of instances, the word influenza is succeeded by the word \u2018virus\u2019, leading to \u2018anti influenza virus<\/strong>\u2019 instead of antibody. However, this is largely speculative, but prompted us to take a more critical look at the manuscript.<\/p>\n\n\n\n One of the major issues we identified were what appeared to be AI hallucinations. This ranged from fabricated conclusions to instances where the citations included were for a different topic entirely. As mentioned above, this can happen, and is a rather documented shortcoming of LLMs. The reason for concern, however, is in the generative AI declaration statement, the authors claimed no generative AI was used in the writing of the review. Although there is no way to conclusively state LLMs were used, the alternative where a human deliberately hallucinated some of these citations would be rather worse. Hence, for the purposes of this blogpost we will assume the inconsistencies are due to AI-mishandling, rather than human failure. Two of the major categories of hallucinations are highlighted below:<\/p>\n\n\n\n Inaccurate citations<\/strong><\/p>\n\n\n\n There are several inaccurate citations in the manuscript. We detail a few below and attempt to understand how these hallucinations originate. For instance, in Section 6.2, the authors cite Clementi et al., 2011 <\/em>when describing the potential for CF-404 as a localised lung delivery mAb. Clementi et al., 2011 <\/em>has no mention of CF-404, and instead focuses on an entirely different anti-influenza mAb PN-SIA28 6 <\/sup>. Here, it may be that the model saw keywords such as \u2018anti-influenza mAb\u2019 and decided this was an appropriate reference. <\/p>\n\n\n\n However, sometimes the rationale behind including a reference is less straightforward. In Section 6.1, the authors cite Nausch<\/em> et al., 2024 <\/em>in their discussions of anti influenza antibodies being studied in different animal models. Nausch et al., 2024 <\/em>is a study on the production of complex proteins in plants 7<\/sup>, without any direct mention of the animal models described or even influenza A. Here, the association is more dubious and we struggled to identify why Nausch et al., 2024 <\/em>was cited for this specific claim. However, other instances where this paper was cited, in relation to antibody manufacturing, were correct. One can hypothesize that an LLM saw this paper being cited several times in the review when discussing mAbs and considered it a generally correct reference for multiple generic claims.<\/p>\n\n\n\n In the same section the authors state \u2018The CR9114 exhibits broad protection against multiple influenza strains while CR6261 demonstrated strong efficacy against group 1 influenza A subtypes (Chung et al., 2023; Grandea et al., 2010)<\/em>\u2019, but again, neither of the cited sources mention CR9114 or CR6261. It should be noted that the claim made is factually correct 8<\/sup>. <\/p>\n\n\n\n Table 5 from the manuscript summarises findings related to five key antibodies against influenza A. But, some of these seem to be unfounded or at the very least incorrectly referenced. For CR9114, the cited reference by Chung et al. 2023 <\/em>9<\/sup> does not even reference the antibody. The article instead is about monoclonal antibody (mAb) delivery systems, specifically those utilising synthetic nucleic acid delivery. The only passing mention of \u2018influenza\u2019 in the article is about how DNA-based mAb delivery systems have been demonstrated against several infectious diseases, including influenza. Importantly, CR9114 is not administered using a DNA-based delivery system, rather as a purified biologic 8,10,11 <\/sup>.<\/p>\n\n\n\n Misinformation<\/strong><\/p>\n\n\n\n In addition to faulty citations, the authors also make several false claims in the same section about CR9114 and CR6261. They state several adverse effects and highlight localised pain at the site of injection, citing Beukenhorst et al, 2022<\/em> and Phillips et al., 2021<\/em>. This is surprising given CR9114 has never been injected in humans, with all reported human administrations being intranasal 10<\/sup>. <\/p>\n\n\n\n It is difficult to deduce how \u2018localised pain at injection site\u2019 can be inferred. Beukenhorst et al, 2022<\/em> is a literature review focusing on the molecular basis of protection of CR9114, and is not a human study. Regardless, the paper made no mention of localised pain being associated with CR9114 administration 8<\/sup> . Phillips et al., 2022 <\/em>is an in vitro high throughput study on the evolution of CR9114\u2019s breadth of binding and contains no mouse or human experiments. <\/p>\n\n\n\n It should be stated that in a Phase-1 and preclinical study reported by Beukenhorst et al, 2026 <\/em>10<\/sup>, the authors demonstrate good safety profiles in humans across dosing regimens when administered intranasally. <\/p>\n\n\n\n Most of the antibodies highlighted in the table do not have statistically significant efficacy in humans, making it a dangerous mischaracterisation. For CR8020, the study population consisted of healthy individuals subsequently given an influenza challenge, not \u2018infected and hospitalized patients\u2019 according to the clinical trial overview 12<\/sup>, which the authors correctly identify. However, the US ClinicalTrials.gov<\/a> page does not report results of the study. Further inspection of the EU Clinical Trials Register<\/a> shows the trial indicated no statistically significant protection over a placebo. Therefore, the claim it resulted in an \u201880% viral load\/ symptom reduction\u2019 is entirely false, and potentially dangerous.<\/p>\n\n\n\n Wrongful citations are detrimental in a number of ways, however their impacts are especially pronounced when reporting about drugs in clinical trials. In instances where widely accepted facts are stated correctly but improperly cited, though detrimental, it does not explicitly hold the scientific community back. Gross misrepresentations of the adverse effects of some of these drugs, such as the claim that CR9114 results in \u2018injection site reactions\u2019 despite never being injected in humans can wrongfully impact public perception or even interfere with further academic investigations. <\/p>\n\n\n\n This review is one of countless examples highlighting a growing problem in academic publishing: lack of accountability from authors and <\/em><\/strong>reviewers\/editors. The onus of academic integrity falls first on the authors to ensure they are not deliberately, or even unintentionally, misleading the scientific community. <\/p>\n\n\n\n Responsibility falls on reviewers, as well as the editors of the publishing venue, to check. Mistakes can happen, and things can fall through the cracks, but it is still disappointing to see nonetheless. In the case of review articles, the effects are particularly pronounced as they often act as the first point of reference for academics to gain an overview of the field.\u00a0In terms of the paper described here, our research partners have been in touch with the research integrity desk at Frontiers<\/em> and have raised these concerns.<\/p>\n\n\n\n AI has taken away several key barriers to academic writing and when used well is a powerful tool 13<\/sup> , but it shouldn\u2019t also take away our academic integrity. <\/p>\n\n\n\n References<\/p>\n\n\n\n 1. NWS, V. Ghent University rector Petra De Sutter uses fabricated quotes in speech, AI gets blame | VRT NWS: news. VRTNWS <\/em>https:\/\/www.vrt.be\/vrtnws\/en\/2026\/01\/08\/ghent-university-rector-petra-de-sutter-uses-fabricated-quotes-i\/ (2026). <\/p>\n\n\n\n 2. Cheng, A., Calhoun, A. & Reedy, G. Artificial intelligence-assisted academic writing: recommendations for ethical use. Adv. Simul. <\/em>10<\/strong>, 22 (2025). <\/p>\n\n\n\n 3. Xu, Z. et al.<\/em> GhostCite: A Large-Scale Analysis of Citation Validity in the Age of Large Language Models. Preprint at https:\/\/doi.org\/10.48550\/arXiv.2602.06718 (2026). <\/p>\n\n\n\n 4. Alkaissi, H., McFarlane, S. I., Alkaissi, H. & McFarlane, S. I. Artificial Hallucinations in ChatGPT: Implications in Scientific Writing. Cureus <\/em>15<\/strong>, (2023). <\/p>\n\n\n\n 5. Mahrous, N. N. et al.<\/em> Broadly neutralizing monoclonal antibodies against influenza A viruses: current insights and future directions. Front. Microbiol. <\/em>16<\/strong>, (2026). <\/p>\n\n\n\n 6. Clementi, N. et al.<\/em> A Human Monoclonal Antibody with Neutralizing Activity against Highly Divergent Influenza Subtypes. PLOS ONE <\/em>6<\/strong>, e28001 (2011). <\/p>\n\n\n\n 7. Nausch, H., Kn\u00f6dler, M. & Buyel, J. F. Production of Complex Proteins in Plants: From Farming to Manufacturing. in Biopharmaceutical Manufacturing: Progress, Trends and Challenges<\/em> (ed. P\u00f6rtner, R.) 241\u2013278 (Springer International Publishing, Cham, 2023). doi:10.1007\/978-3-031-45669-5_8. <\/p>\n\n\n\n 8. Beukenhorst, A. L. et al.<\/em> The influenza hemagglutinin stem antibody CR9114: Evidence for a narrow evolutionary path towards universal protection. Front. Virol. <\/em>2<\/strong>, (2022). <\/p>\n\n\n\n 9. Chung, C. et al.<\/em> Expanding the Reach of Monoclonal Antibodies: A Review of Synthetic Nucleic Acid Delivery in Immunotherapy. Antibodies <\/em>12<\/strong>, 46 (2023). <\/p>\n\n\n\n 10. Beukenhorst, A. L. et al.<\/em> Phase 1 and preclinical studies reveal safety, pharmacokinetics, and efficacy of intranasal delivery of the influenza antibody CR9114. Sci. Transl. Med. <\/em>18<\/strong>, eadz1580 (2026). <\/p>\n\n\n\n 11. Beukenhorst, A. L. et al.<\/em> Intranasal administration of a panreactive influenza antibody reveals Fc-independent mode of protection. Sci. Rep. <\/em>15<\/strong>, 10309 (2025). <\/p>\n\n\n\n 12. Crucell Holland BV. Randomised, Double-Blind, Placebo-Controlled, Phase IIa Study in Healthy Volunteers to Evaluate the Protective Efficacy and Safety of CR8020 in an Influenza Challenge Model<\/em>. https:\/\/clinicaltrials.gov\/study\/NCT01938352 (2019). <\/p>\n\n\n\n 13. Boustane, H., Chroqui, R. & Sabil, A. The impact of artificial intelligence on academic writing: a systematic literature review. Int. J. Learn. Technol. <\/em>20<\/strong>, 316\u2013343 (2025). <\/p>\n","protected":false},"excerpt":{"rendered":" In the last five or so years, large language models (LLMs) have transformed from a novel regurgitator of haphazardly stitched together sentences to an almost \u2018human\u2019 personality standing by our side as we tackle life. Whilst the perceived humanity of these models is the topic for perhaps a future blogpost, it is almost undeniable to […]<\/p>\n","protected":false},"author":149,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"nf_dc_page":"","wikipediapreview_detectlinks":true,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"ngg_post_thumbnail":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[123,263,48],"tags":[],"ppma_author":[924],"class_list":["post-14157","post","type-post","status-publish","format-standard","hentry","category-commentary","category-news","category-publication"],"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"authors":[{"term_id":924,"user_id":149,"is_guest":0,"slug":"rebonto","display_name":"Rebonto Haque","avatar_url":"https:\/\/secure.gravatar.com\/avatar\/fdd06921411f91130d686e6086353fbd0d4f61196abffbebbbe64a312e3307f3?s=96&d=mm&r=g","0":null,"1":"","2":"","3":"","4":"","5":"","6":"","7":"","8":""}],"_links":{"self":[{"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/posts\/14157","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/users\/149"}],"replies":[{"embeddable":true,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/comments?post=14157"}],"version-history":[{"count":4,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/posts\/14157\/revisions"}],"predecessor-version":[{"id":14161,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/posts\/14157\/revisions\/14161"}],"wp:attachment":[{"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/media?parent=14157"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/categories?post=14157"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/tags?post=14157"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/www.blopig.com\/blog\/wp-json\/wp\/v2\/ppma_author?post=14157"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}