Unlike human intelligence, which arises from complex biological processes, large language models operate through statistical pattern recognition. These systems are probabilistic models trained to predict the most likely sequence of words based on vast datasets. Consequently, they do not truly “understand” concepts; instead, they simulate knowledge by reproducing linguistic patterns. Their outputs are therefore constrained by the quality of training data, algorithmic design, and even commercial or institutional influences that may shape which information is emphasized or omitted. These limitations raise important concerns about intellectual autonomy and the preservation of analytical reasoning in scientific environments.

One of the most widely discussed limitations of large language models is the phenomenon known as “hallucination”. Originally described in research on neural machine translation and later adopted to characterize generative AI systems, hallucination refers to the confident production of factually incorrect or unsupported information [1,2]. Because these models prioritize linguistic plausibility rather than factual verification, they may generate coherent yet erroneous responses instead of acknowledging uncertainty. As a result, outputs may appear convincing and well-structured while containing fabricated details, inaccurate claims, or nonexistent references. In healthcare, such hallucinations extend beyond technical inaccuracies; they can reinforce misinformation, create misplaced confidence in algorithmic authority, and ultimately influence professional judgment, potentially affecting diagnostic reasoning and treatment decisions.

In endodontics, the implications of hallucinated information can be particularly concerning. In the scientific domain, increasing reliance on AI for drafting manuscripts, summarizing literature, or assisting in the development of research protocols introduces the risk of fabricated references, distorted interpretations, or oversimplified representations of complex findings. In this context, manipulated citations generated by AI in endodontic research can distort literature reviews, mislead readers, and compromise the integrity of scientific discussion. Moreover, literature is beginning to witness the emergence of highly speculative conceptual publications proposing novel terminologies, theoretical constructions, or mechanistic explanations that remain entirely unvalidated. When such frameworks are generated or amplified through AI-assisted writing, they may create an illusion of scientific novelty while lacking biological plausibility or empirical support. Without rigorous verification, these ideas may introduce misleading concepts into literature. Once inaccurate information enters the scientific record, it may propagate through secondary citations, gradually contaminating the evidence base and conferring unwarranted legitimacy to incorrect claims. Because academic publications form the foundation for guidelines, education, and clinical decision-making, the introduction of hallucinated content threatens not only individual studies but also the reliability of the broader knowledge framework that supports the discipline. Editors and reviewers must therefore remain particularly vigilant when evaluating manuscripts that may involve AI-assisted writing or AI-generated content.

The potential consequences extend to clinical practice. Endodontic diagnosis relies on the careful integration of clinical findings, radiographic interpretation, and patient history. AI-generated responses that present confident but oversimplified conclusions risk masking uncertainty and discouraging differential diagnosis. Similarly, treatment suggestions generated without appropriate contextual understanding may promote generalized protocols that overlook anatomical variability and patient-specific conditions. When such outputs are accepted without verification, clinical reasoning and ultimately patient care may be compromised.

Addressing these challenges requires more than technological refinement; it demands a culture of critical appraisal. AI-generated information must be interpreted through the lens of professional expertise and verified against primary scientific evidence. Cross-checking claims, maintaining rigorous peer review, and restricting AI to supportive roles—such as information retrieval or hypothesis generation—are essential safeguards.

AI undoubtedly offers important opportunities to enhance access to information, accelerate knowledge exchange, and support education. Yet its limitations, including hallucinations, bias, and the risk of misinformation, require sustained vigilance. Dentistry must therefore engage with AI in a balanced and reflective manner: embracing its benefits while preserving the methodological rigor and critical reasoning that underpin evidence-based practice. Safeguarding the reliability of the scientific record must remain a shared responsibility among authors, reviewers, and editors as AI tools become increasingly integrated into the research ecosystem. Only through such collective vigilance can AI strengthen—rather than undermine—the scientific and clinical foundations of endodontic practice.

Citations to this article as recorded by