Can Large Language Models Act as “Co-Auditors”?

This study explores the integration of large language models (LLMs) into audit workflows as "co-auditors," emphasizing the necessity of embedding them within frameworks that ensure evidence traceability, governance, and human accountability. Despite growing interest in AI-augmented auditing, prior work has not systematically bridged LLM technical capabilities with audit standards and regulatory compliance requirements. Through a narrative literature review synthesizing audit doctrine, AI governance frameworks, and natural language processing research, the study examines how such integration can be achieved. Rather than substituting professional judgment, LLMs offer auditable support that enhances assurance processes. By incorporating hybrid retrieval, policy-constrained generation, and cryptographic provenance, the proposed architecture addresses both factual reliability and regulatory compliance. The findings underscore that effective LLM deployment requires strict alignment with standards. Ultimately, the research confirms that trustworthy AI in auditing depends on robust technical safeguards, governance structures, and sustained human oversight.