LLMs in Audit – A Double-Edged Algorithm

INTRODUCTION

The exuberance associated with artificial intelligence (“AI”) has seamlessly transcended the practice of auditing. Large Language Models (“LLMs”) are heralded as a transformative solution due to their apparent ability to infer and reason both structured and unstructured data. Traditional auditing applications, constrained by rules and structures, are inherently rigid and complex, requiring intricate coding skills to derive substantive insights. In contrast, LLMs appear to be sentient, with their ability to interpret simple natural language instructions. Their ability to perform various tasks, from complex data analysis to code generation, makes them a versatile, unified tool. A simple instruction can now accomplish what previously required multiple applications and data analysis expertise.

This apparent ease of use and accessibility has made LLMs attractive to auditors seeking efficiency and potentially offers smaller audit firms an economical means to bridge their technology gap with larger competitors. As such, it is not surprising that most auditors intend to use LLMs¹. However, the use of LLMs for audits may be fraught with risks, particularly when they are used in relation to matters that involve professional judgement. This article seeks to explore these issues.

1 “Audit Survey 202
Read More