Vertical federated learning (VFL) has become a cutting-edge approach for collaborative machine learning, as it enables multiple organizations to jointly train models on vertically partitioned data where each party holds different features of a common dataset without revealing any of their sensitive data. This is particularly useful for purposes of driving business operations in sectors characterized by strict data confidentiality laws such as banking, health care, and telecommunications. Nevertheless, VFL comes with privacy and security issues that are more complex than one exists in conventional or horizontal federated learning systems. In this paper, we present an extensive review of these challenges, focusing consideration on the significance of communications protocols that serve for safeguarding information and model confidentiality. Different protocols like secure multiparty computation (SMPC), homomorphic encryption (HE), as well as differential privacy (DP), their virtues and demerits, and their influence on the level of security and communication are discussed. Moreover, we explore and examine about how existing advancements of VFL can help to improve the efficiency of model training, thus offering path for secure and scalable solutions to existing challenges. By identifying current gaps and reviewing effective solutions, this paper aims to guide future research in building VFL frameworks that are both secure and efficient.