Skip to main content

GenAI for tax, legal, and audit functions

Don’t compromise trust, quality, controls, or confidence for potential efficiency savings

The use of Generative AI (GenAI) in tax, legal, and audit functions can bring sizable benefits, but it is crucial to establish a clear framework for ensuring trust, quality, control, and confidence.

With the use of GenAI, tax, legal and audit teams can access information more easily, often with greater precision. It can automate repetitive tasks such as generating first-draft documents, research and development tax-relief identification and expense categorization for tax purposes. However, it's important to weigh the human effort required to validate or fact-check the AI system's output and understand its potential effect on existing governance, systems, and controls.

To reap the rewards of GenAI while also managing its risks, a clear, rigorous framework is needed for ensuring trust, quality, control, and confidence.



Legal, tax, and finance professionals handle sensitive legal and financial data, making it challenging to use GenAI models due to security, confidentiality, and privacy concerns. Navigating these issues requires trust in the people and systems, and is crucial at every stage of the workflow lifecycle. The AI model must be secure to prevent data breaches, and AI-produced results must be accurate, reliable, and undergo human validation.

GenAI can provide rapid, high-quality insights, but human intelligence is still necessary to verify output and prevent errors.



GenAI can be a valuable tool for businesses, but it is important to use it carefully to avoid errors that could result in fines or damage to reputation. Risks include AI "hallucinations," legal issues such as plagiarism and copyright infringement, privacy and data ownership challenges, and lack of transparency and accountability. When used effectively, GenAI can provide rapid, high-quality insights in areas such as tax (e.g., significantly reduce manual transaction processing) and forensic accounting (e.g., identifiying and cleaning questionable transactions, data anomalies, and missing data fields), but human intelligence is still necessary to verify output and prevent errors.



Lawyers, auditors, and tax specialists must maintain high ethical standards and professional judgement when using AI. Human oversight is crucial to prevent AI from inventing non-existent reference cases and exacerbating systemic bias. Ongoing training and rigorous safeguards are necessary to reinforce sound judgement, protect sensitive data, and comply with regulatory rules. Businesses should confirm that AI models are working as designed and understand the data sets being used, permissions required, and where data and outputs are shared. Organizations can protect data used in tax, legal, and audit workflows by building a structured tool for massive storage and fragmenting it using criteria-based permits.

The fundamental challenge here is capitalizing on AI’s power without losing control.



Businesses are calling for increased global collaboration around AI regulation and development, with 78% of leaders surveyed by Deloitte between October and December 2023 stating that more governmental regulation of AI is needed. 72% also believe that there is currently not enough global collaboration to ensure responsible development of AI-powered systems. This highlights the understanding that the unique and complex risks associated with AI may be too powerful and far-reaching for individual organizations to regulate themselves. AI can be used by legal professionals or compliance officers to speed up manual processes, but detailed human review is crucial for more sensitive or nuanced information.

In summary, GenAI has huge potential benefits, but it comes with risks that could harm brand, reputation, customer trust, or legal compliance. Human oversight, risk assessment, and creativity are, and will remain, essential components of legal, tax, and finance functions—helping to ensure and maintain accuracy, confidentiality, trust, and control.

Did you find this useful?

Thanks for your feedback

If you would like to help improve further, please complete a 3-minute survey