Somewhere along my academic journey, through the long and sometimes exhausting procession of coursework, research proposals, journal articles, theses, dissertations, methodological arguments, statistical analyses, corrections, examinations, defences and the relentless search for credible authorities, I encountered an article whose title immediately arrested my attention: He Liu's “Fabricated citations in the age of AI: A wake-up call for editors, reviewers, and authors.” Published in the Journal of Dental Sciences, Volume 21, Issue 1, pages 679–680, with DOI 10.1016/j.jds.2025.10.024, the correspondence initially appeared to me as another contribution to the rapidly expanding conversation about artificial intelligence and scholarly writing. Yet the more closely I considered its account, the more uncomfortable its implications became. The article was ostensibly about fabricated references, but beneath that immediate problem lay something considerably larger and older than artificial intelligence itself: the possibility that academic honesty can drift, quietly, incrementally and almost imperceptibly, until practices that should trouble the scholarly conscience begin to acquire the familiarity of normal academic life.
The scriptural language of honesty provides an unexpectedly powerful entrance into this discussion. In 1 Timothy 2:1–2, Paul begins with a remarkable ordering: “first of all” come supplications, prayers, intercessions and thanksgivings for all people, then for kings and all who are in authority, toward the end that people may lead “a quiet and peaceable life in all godliness and honesty.” The King James Version's “honesty” carries the sense that several modern translations render as dignity: gravity, honourableness, seriousness of character and conduct worthy of respect. That conception reaches considerably deeper than simply avoiding an outright lie. It speaks to the moral condition in which truth can safely live. Scholarship requires precisely such an atmosphere. Academic honesty is not merely the absence of plagiarism detected by software. It is a disposition toward truth governing what we claim to know, how we came to know it, whose intellectual labour we have used, what evidence supports our assertions, what contribution we ourselves made, what uncertainties remain and where the boundaries of our knowledge properly lie. Strip scholarship of that honesty and universities may remain physically magnificent, libraries may remain full, graduations may continue with pomp and circumstance, professorial titles may multiply and publication counts may soar, while the intellectual foundations beneath the whole enterprise quietly begin to move.
Liu's account is therefore worth examining with almost forensic attention. Writing from his experience as an academic editor, editorial board member and reviewer, he describes encountering a manuscript containing serious methodological flaws. Before making the rejection decision, he examined its references more closely and discovered that nearly 80 per cent were fabricated or nonexistent. More troubling still, he reports that the manuscript had already passed the journal's initial technical screening and undergone three rounds of peer review by three different reviewers, without the editorial staff or reviewers detecting that most of the references were false. Consider the anatomy of that failure. A manuscript was prepared. References were inserted. It was submitted. Technical screening was traversed. Reviewers examined it. Revisions apparently proceeded. Further review followed. Yet a bibliography reportedly dominated by nonexistent sources travelled astonishingly far through machinery specifically established to scrutinise scholarly work. Artificial intelligence may have manufactured the smoke, but several human sentries apparently watched it drift through the gates.
That is what makes the incident more consequential than another cautionary tale about artificial intelligence. The obvious question is whether AI fabricated the references. The more disturbing question is: how did scholarly systems fail to detect them? If a nonexistent publication can survive authorship, submission, technical screening and repeated peer review before somebody finally asks whether it actually exists, then the problem cannot be located exclusively inside the machine. It also resides within habits of academic trust, workloads, assumptions, incentives and perhaps an increasing tendency to mistake the outward appearance of scholarship for scholarship itself. The machine may hallucinate; but only human beings can decide whether to verify the hallucination before allowing it into the scholarly record.
There is something almost uncanny about the modern fabricated reference because it need not look fabricated at all. It may carry the names of plausible authors, an intellectually sophisticated title, the name of a genuine journal, a credible year of publication, an appropriate volume and issue, convincing page numbers and something shaped perfectly like a Digital Object Identifier. It may arrive alphabetised, punctuated and formatted impeccably in APA style. It may sit quietly among genuine authorities, borrowing legitimacy from its neighbours and inviting the hurried reader to move along. There is only one inconvenient problem: the publication may never have existed. Nobody designed the study. Nobody recruited the participants. Nobody collected the data. Nobody performed the analysis. Nobody wrote the conclusions. No journal published them. Yet there stands the citation, dressed in the ceremonial garments of scholarship and demanding to be received as evidence.
This exposes one of the profound epistemological challenges presented by generative artificial intelligence: plausibility is not truth. A large language model can learn the linguistic architecture of scholarship with extraordinary sophistication. It recognises how academic titles sound, how journal references are structured, how abstracts unfold, how arguments are qualified and how bibliographies are punctuated. Consequently, it may sometimes reproduce the outward architecture of scholarship without possessing the evidential foundations upon which scholarship depends. It can construct the façade without the building behind it, the label without the substance, the map without the territory. Proverbs 14:15 could scarcely be more contemporary: “The simple believeth every word: but the prudent man looketh well to his going.” In the age of generative AI, prudence has become not merely a virtue but an academic methodology.
Yet we would be intellectually dishonest ourselves if we conveniently made artificial intelligence the scapegoat for sins that preceded it by centuries. Machines did not invent plagiarism. They did not invent fabrication of evidence, manipulation of data, selective reporting, appropriation of ideas, suppression of inconvenient findings or undeserved academic credit. Long before a generative model could hallucinate an article, human beings were perfectly capable of citing authorities they had never read and presenting intellectual labour as their own. Artificial intelligence did not create the temptation to circumvent truth. What it has done is give some ancient temptations astonishing new speed, scale, polish and plausibility. What once required considerable effort can sometimes now be manufactured in seconds and delivered in prose so polished that the reader's critical faculties may be seduced precisely by the fluency of the error.
This reflection inevitably brought back another phenomenon I have encountered in real academic life, one requiring no artificial intelligence whatsoever. There are occasions when someone asks to be included as an author of an academic article to which he or she has made no substantive intellectual contribution. Sometimes the request arrives almost casually: include my name. Sometimes friendship provides the atmosphere. Sometimes professional relationships make refusal uncomfortable. Sometimes hierarchy may hover silently in the room. Sometimes authorship appears to be treated almost as an academic favour that one colleague can extend to another. Yet after friendship, hierarchy, convenience and collegiality have been stripped away, one uncompromising question remains: what exactly does the name of an author certify?
Authorship is not decoration. It is not an academic Christmas card upon which additional names may be inscribed because goodwill exists between the parties. It is not a souvenir, honorary ornament, transferable favour or token of friendship. Authorship confers credit, but it simultaneously carries responsibility. This is not merely a philosophical position. The International Committee of Medical Journal Editors states explicitly that authorship has academic, social and financial implications and also entails responsibility and accountability. Its current criteria require substantial contribution to the conception or design of the work, or to acquisition, analysis or interpretation of data; participation in drafting or critical intellectual review; approval of the final version; and agreement to be accountable for the integrity of the work. Those who do not satisfy the authorship criteria should be acknowledged rather than represented as authors.
Modern scholarship has gone even further in trying to make contribution visible. The CRediT Contributor Role Taxonomy, now formalised as an ANSI/NISO standard, identifies fourteen categories through which contributions to research can be transparently described, including conceptualisation, methodology, investigation, formal analysis, data curation, validation, software, visualisation and writing. The significance is profound. Scholarship is moving toward a world in which the question should increasingly cease to be merely “Whose names appear on this paper?” and become “What did each person actually contribute?” That is not bureaucratic pedantry. It is the restoration of correspondence between academic representation and academic reality.
Here, remarkably, the fabricated citation and unearned authorship meet upon precisely the same ethical terrain. In one case, a reference appears without the scholarship behind it; in the other, an author appears without the contribution behind the name. One manufactures authority for evidence; the other manufactures credit for intellectual labour. Their mechanisms differ, but their internal pathology is strikingly similar: appearance has become detached from reality. The citation looks authentic but may have no source. The authorship looks earned but may have no contribution. The form remains immaculate while the substance has quietly disappeared. Proverbs 12:22 therefore speaks with almost surgical precision: “Lying lips are abomination to the LORD: but they that deal truly are his delight.” For the scholar, those final words constitute an entire ethical charter: deal truly.
But our forensic examination must proceed still deeper, because establishing that a reference exists does not complete the scholarly duty. A genuine reference can still become a false witness. An article may exist exactly as cited. Its DOI may resolve. Its authors may be correctly named. The volume, issue and pages may be flawless. Yet the article may not support the proposition beside which it has been placed. A correlation quietly metamorphoses into causation. Findings from one population are transported into another as though geography, context and sampling were irrelevant. A tentative association becomes an established effect. An author's cautious “may” becomes another writer's confident “does.” Limitations disappear. Qualifications evaporate. A secondary citation masquerades as engagement with the primary source. The reference is real, but the testimony extracted from it is false.
The forensic scholar must therefore interrogate evidence through successive gates. Does the source exist? Are its bibliographic particulars accurate? Has the source actually been consulted? Does it say what the writer claims it says? Does the methodology justify that conclusion? Does the sample permit that generalisation? Have limitations been preserved? Has uncertainty been honestly represented? Only after these questions have been answered can a citation properly perform its evidential function. First Thessalonians 5:21 provides a wonderfully economical scholarly ethic: “Prove all things; hold fast that which is good.” Test. Examine. Verify. Then retain.
The Bereans in Acts 17:11 provide an even richer intellectual model. They received what they heard “with all readiness of mind,” yet searched the Scriptures daily to determine “whether those things were so.” Notice the magnificent equilibrium: openness without gullibility; readiness without surrender; receptivity followed by verification. That is precisely the intellectual disposition the university requires toward artificial intelligence. We need neither prostrate ourselves before the machine nor flee from it in technological panic. Receive what is useful. Examine what is claimed. Search whether these things are so. The Berean instinct may prove one of the finest epistemological disciplines for the AI university.
For students, this distinction is decisive. Generative AI can be a formidable educational instrument. A learner struggling with an unfamiliar concept can request an explanation, reject the first explanation as too technical, ask for an analogy, demand a worked example, interrogate an assumption, compare theoretical perspectives and continue the conversation until comprehension begins to emerge. A student can use AI to expose weaknesses in an argument, formulate questions for further research, clarify terminology, compare methodologies or challenge an interpretation. Used this way, artificial intelligence extends intellectual reach. It becomes scaffolding around learning rather than a replacement for learning.
But there is another road, broad, smooth and dangerously attractive. A student may request a complete literature review with thirty recent scholarly references, wait several seconds, copy the output and submit it. The headings may be excellent. The prose may roll magnificently. Citations may march confidently across every page. A handsome reference list may appear at the end. Yet the student may not have located a single paper, interrogated a single methodology, examined a single dataset or understood the argument supposedly bearing his or her name. The document has expanded dramatically while the scholar has remained intellectually stationary. The appearance of learning has increased while learning itself may have diminished. We would then have produced the strangest of academic artefacts: a sophisticated paper attached to an unsophisticated understanding.
This is why the question confronting universities can no longer simply be, “Did you use AI?” That question is rapidly becoming too crude. Artificial intelligence is entering search engines, word processors, research platforms, statistical environments, coding tools and professional software. The more searching question is: what intellectual work did you personally perform? Can you defend the argument? Can you locate the source? Can you explain why that methodology was appropriate? Can you distinguish evidence from interpretation? Can you reproduce the calculation? Can you identify the strongest limitation? Can you explain why an alternative interpretation was rejected? Can you demonstrate that the conclusion emerged from understanding rather than merely materialising upon a screen?
In such an environment, the viva voce acquires renewed importance. A thesis may be polished by software. Grammar may be corrected. Sentences may be reorganised. Statistical outputs may be translated into elegant prose. References may be reformatted. But sustained oral interrogation remains remarkably capable of exposing the distance between possessing a document and possessing knowledge. Why did you choose this theoretical framework? What exactly did this author establish? Why was this statistical test appropriate? What assumptions underpin this model? What would happen if that assumption changed? Where precisely in the source is the evidence for this assertion? What is the strongest argument against your conclusion? These questions penetrate beneath the typography of the thesis and encounter the mind claiming authorship of it.
For a technical university such as the Technical University of Mombasa, this conversation should carry particular force because technical disciplines possess an unforgiving relationship with reality. Reality audits us. A building cannot be persuaded by eloquent prose to remain standing. Quantities cannot be argued into reconciliation. Concrete does not acquire compressive strength because its test certificate looks impressive. A cost plan does not balance because its spreadsheet possesses beautiful formatting. An electrical installation does not become safe because its designer has written a magnificent explanation. Software does not execute because its documentation sounds authoritative. A laboratory experiment does not become reproducible because its bibliography is long. Sooner or later the structure, measurement, calculation, experiment, design or code collides with reality, and reality is wonderfully unimpressed by academic cosmetics.
Therein lies an extraordinary opportunity for a technical university. The same intellectual culture that insists that measurements reconcile, calculations withstand checking, designs satisfy constraints, experiments remain reproducible and materials satisfy specifications can be carried deliberately into scholarship. Claims must reconcile with evidence. Citations must reconcile with sources. Authorship must reconcile with contribution. Conclusions must reconcile with methodology. Assertions must reconcile with data. Academic representation must reconcile with academic reality. Proverbs 11:1 could almost be engraved upon the entrance to that intellectual workshop: “A false balance is abomination to the LORD: but a just weight is his delight.” The ancient marketplace and the modern university are separated by millennia, but their ethical architecture remains recognisable. The merchant's dishonest balance attributed weight that the commodity did not possess. Dishonest scholarship attributes evidential or intellectual weight that the work does not possess. In both cases, the moral failure lies in misrepresentation of measure.
Yet perhaps the greatest danger is not spectacular academic fraud. It is drift. Integrity rarely disappears in one dramatic collapse. Standards migrate by millimetres. One citation is inserted without being checked. One source is cited without being read. One name is added to an article as a favour. One inconvenient limitation is quietly omitted because it weakens the argument. One student's work is rewritten so extensively that supervision begins imperceptibly becoming authorship. One reviewer assumes technical screening checked the references. The technical editor assumes reviewers will check them. The editor assumes the authors did. Everyone assumes; nobody verifies. The questionable becomes familiar, the familiar becomes normal, the normal becomes custom, and custom eventually hardens into culture.
“A little leaven leaveneth the whole lump,” Paul warns in Galatians 5:9. The metaphor is almost forensic in its description of ethical drift. Leaven does not transform dough through a thunderclap. It works quietly, invisibly and progressively until what began locally has permeated the whole. Academic dishonesty can move in precisely this manner. One apparently harmless accommodation becomes precedent. Precedent becomes habit. Habit becomes expectation. Expectation becomes culture. And eventually culture begins disciplining the very people who resist it. The scholar who insists upon proper attribution becomes “difficult.” The researcher who refuses honorary authorship becomes “uncooperative.” The examiner who interrogates evidence becomes “too strict.” The person maintaining the standard begins to appear unreasonable precisely because drift has moved everybody else.
That is why academic honesty cannot be reduced to plagiarism software, examination regulations or declarations signed during submission. Integrity is an ecosystem. Students inhabit it, but so do lecturers. Researchers inhabit it, but so do supervisors. Authors inhabit it, but so do editors and reviewers. Junior academics inhabit it, but professors and institutional leaders do not somehow float above it. Academic honesty has no academic rank. Indeed, the greater one's scholarly authority, the greater the responsibility to ensure that authority does not become a mechanism through which undeserved credit, weak evidence or convenient exceptions acquire legitimacy.
Nor should the response become technological panic. Artificial intelligence is too useful to be dismissed as an enemy and too fallible to be enthroned as an oracle. The wiser university must inhabit the demanding territory between worship and prohibition. Use AI where appropriate. Understand its capabilities. Learn its limitations. Disclose its use where required. Interrogate its outputs. Verify its references. Recalculate important results. Return to primary sources. Preserve professional judgment precisely where responsibility demands it. Significantly, current ICMJE guidance itself addresses AI-assisted technologies and requires appropriate disclosure while retaining human responsibility for the submitted work. The direction of serious scholarship is therefore not toward pretending AI does not exist, but toward ensuring that technological assistance never dissolves human accountability.
There is something almost providential about this historical moment. One of humanity's newest technologies is compelling the university to rediscover one of scholarship's oldest virtues: disciplined scepticism. Not cynicism. Not technophobia. Not suspicion for suspicion's sake. Rather, the ancient intellectual habit of asking: What is the evidence? A citation appears: locate it. A DOI appears: resolve it. A statistic appears: trace it. A quotation appears: inspect its context. A causal claim appears: interrogate the research design. An authorship claim appears: identify the contribution. An AI-generated paragraph appears in immaculate prose: appreciate its fluency if one wishes, but examine its truth.
Perhaps, therefore, the most dangerous sentence in the AI university is not “I do not know.” Those four words remain among the most intellectually honourable words a scholar can utter because they truthfully mark the frontier between knowledge and ignorance. The more dangerous sentence is “It looks right.” Looking right is not being right. Sounding scholarly is not scholarship. Having citations is not having evidence. Possessing a thesis is not necessarily possessing the knowledge contained within it. Appearing upon an author line is not the same thing as having earned authorship. Formatting is not authentication. Fluency is not truth. Confidence is not evidence.
Paul's exhortation in Philippians 4:8 gathers these concerns into a magnificent ethical compass: “Whatsoever things are true, whatsoever things are honest, whatsoever things are just... think on these things.” True. Honest. Just. Truth demands correspondence between claim and reality. Honesty demands integrity in representing that reality. Justice demands that intellectual credit be given where it is due and withheld where it is not. These are not merely devotional virtues belonging to private spirituality. They describe with astonishing precision the moral foundations upon which responsible scholarship rests.
Academic honesty must therefore be protected not merely from spectacular misconduct, but from drift: from the little accommodations through which the unacceptable becomes familiar, the questionable becomes normal and eventually the dishonest becomes respectable. Universities rarely wake up one morning and collectively decide that truth no longer matters. Standards are more likely to be surrendered incrementally, through countless seemingly insignificant decisions to assume rather than verify, appropriate rather than attribute, conceal rather than disclose, embellish rather than qualify, copy rather than comprehend, and confer authorship rather than require contribution.
Artificial intelligence did not create this moral crossroads. It has illuminated it with extraordinary brightness and dramatically increased the speed at which we can travel in either direction. The machine can help us search farther, compare faster, analyse more efficiently, communicate more clearly and interrogate ideas more deeply than earlier generations could easily imagine. Used wisely, it can enlarge the intellectual capacities of the university. Used carelessly, it can manufacture scholarly appearance at unprecedented scale. The technology is powerful; the moral choice remains human.
The governing principle for the university in this age need therefore not be complicated. The machine may suggest; the scholar must verify. The machine may generate; the scholar must interrogate. The machine may calculate; the professional must understand. The machine may cite; the researcher must locate. The machine may write persuasively; the author must remain accountable. The institution may confer degrees and titles; integrity alone confers credibility.
Somewhere tonight, another immaculate bibliography may be materialising upon a screen. Somewhere, another magnificent paragraph may be appearing before someone who has not read the literature beneath it. Somewhere, a perfectly genuine article may be enlisted to support a proposition its authors never made. Somewhere, perhaps, another person may be asking for a name to be added to research to which that person did not contribute. Each incident may appear small when viewed alone. Together they reveal one of the defining ethical challenges confronting contemporary scholarship.
And here Scripture turns the searchlight finally away from the machine and toward us. “Examine yourselves,” Paul writes in 2 Corinthians 13:5. That may be the most uncomfortable and necessary instruction of all. The academy must examine artificial intelligence, certainly, but it must also examine its own habits, incentives, publication pressures, authorship cultures, supervisory practices, examination systems, reviewing routines and institutional consciences. For academic honesty cannot be protected merely by watching machines. It must be protected by scholars committed to truth even when nobody is checking.
When the reference does not exist, we must say so. When the evidence does not support the claim, we must say so. When the contribution does not exist, the authorship should not exist either.
That is not hostility to artificial intelligence. It is not resistance to technological progress. It is something considerably older and considerably more important.
It is fidelity to scholarship.
And in an age when appearances can be manufactured almost effortlessly, when fluent prose can outrun understanding, when citations can be generated faster than they can be read, and when academic credit can sometimes become detached from intellectual contribution, the university has an obligation that no algorithm can discharge on its behalf: to keep academic honesty firmly anchored while everything around it threatens to drift.