This study aims to examine the application of next-generation artificial intelligence models, particularly large language models (LLMs), in financial risk management and accounting. With the rapid development of generative artificial intelligence, models such as GPT and BERT have created new capabilities for processing structured and unstructured data, extracting financial information, predicting risks, and supporting decision-making. Using a mixed-method approach, including a systematic review of the literature, comparative analysis of models, and examination of practical applications, this study investigates the role of LLMs in areas such as **credit risk, regulatory compliance, feature selection, economic simulation, financial reporting, and anomaly detection**. The findings indicate that LLMs, through their ability to extract information from financial texts and integrate diverse types of data, can serve as effective complements to traditional financial risk management models. Furthermore, the use of specialized models fine-tuned with proprietary data, the integration of LLMs with traditional methods, and the application of synthetic data can enhance the potential of this technology in financial applications. However, challenges such as training data quality, algorithmic bias, interpretability, privacy, regulatory requirements, and dependence on proprietary data represent significant limitations to the widespread adoption of these models. Based on the findings, the use of LLMs as a complete replacement for traditional methods is not recommended. Instead, a **hybrid approach combining LLMs, quantitative models, and human oversight** can provide a more appropriate framework for achieving more accurate, transparent, and reliable decision-making in financial risk management.