Biomedical Named Entity Recognition (BioNER) is vital for extracting structured information from vast amounts of unstructured biomedical text. This study introduces an enhanced BioNER approach by fine-tuning the BioBERT model using adversarial training techniques—specifically, Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD)—to bolster model robustness against input perturbations. By incorporating optimized token alignment strategies, the proposed method significantly improves the identification and classification of biomedical entities across multiple benchmark datasets, including MedMentions, BC5CDR, and i2b2 2010. Comprehensive evaluations using metrics such as Precision, Recall, F1-score, and Entity-Level Accuracy demonstrate that the model consistently surpasses current state-of-the-art systems. This work not only highlights the advantages of adversarial training for domain-specific language models but also sets a new standard for robust and accurate biomedical NER systems.