Artificial intelligence and research integrity in reproductive medicine: reproducibility, transparency, accountability, and responsible translation

Abstract Artificial intelligence (AI) is increasingly being investigated across reproductive medicine, including semen analysis, sperm detection and selection, oocyte and embryo assessment, prediction of assisted reproductive technology outcomes, clinical decision support, and integration of multi-omics data. Large language models and generative AI are also emerging as tools for scientific writing, education, documentation, and research support. This review critically examines the current applications, clinical readiness, and research-integrity implications of AI in reproductive medicine. Although many models demonstrate promising performance in classification, ranking, or outcome prediction, the available evidence remains predominantly retrospective, task-specific, and derived from single-center or internally validated datasets. High technical performance does not necessarily translate into improved cumulative live birth, patient safety, cost-effectiveness, workflow efficiency, or equitable clinical outcomes. Important barriers include dataset heterogeneity, limited external and prospective validation, inadequate calibration assessment, data leakage, algorithmic bias, limited interpretability, privacy and cybersecurity risks, and vulnerability to performance drift. Generative AI introduces additional concerns, including fabricated references, inaccurate scientific summaries, undisclosed contributions, confidentiality breaches, and the generation or alteration of images that may be misrepresented as empirical evidence. Responsible translation therefore requires transparent reporting, independent temporal and geographic validation, subgroup and fairness analyses, prospective clinical-impact evaluation, preservation of data and model provenance, and continuous post-deployment surveillance. Current international guidance further emphasizes that AI systems cannot qualify as authors and that accountable human researchers must verify generated content and disclose material AI use. AI should presently be regarded as an assistive technology that augments, rather than replaces, embryologists, clinicians, and researchers. Future progress will depend on multicenter prospective studies using patient-important outcomes, design-appropriate reporting frameworks, risk-based regulatory oversight, and clearly defined human accountability throughout the AI lifecycle.