Large language models represent a major advance in artificial intelligence research. These models are based on complex neural networks trained on large datasets and can provide unique capabilities in various domains. The application of large language models in the audit field has also created challenges in performing repetitive tasks, evaluating internal controls, providing procedures, and planning audits. Studies have also defined AI auditing as a mechanism that ensures that AI systems are ethically, legally, and technically sound. However, existing auditing practices have failed to address the challenges posed by large language models that exhibit emerging capabilities. The present study, by reviewing the existing literature, while defining and classifying these models, has raised the opportunities and challenges of using large language models in the field of auditing and has presented a three-level framework for their auditing. The results of this study provide researchers with a structured basis for classifying large language models. The study also shows that audits, when conducted in a structured and coordinated manner at all three levels, can be a practical mechanism for managing some of the ethical and social risks posed by large language models.