Since advanced artificial intelligence techniques, such as machine learning algorithms (MLA) and natural language processing (NLP), are moving automation to sophisticated levels not seen before, accountants need to “concentrate on areas which remain difficult to automate, such as where human judgement or a deep understanding of the business environment is required, or where tasks depend on the knowledge and application of highly complex rules” (Institute of Chartered Accountants in England and Wales). In addition to the current level of business familiarity, the new focus will require stronger technical and statistical skills, which in the past may have been delegated to technology experts. In turn, firms will need to respond by offering the resources and support for staff to attain the requisite skill levels.
The accounting profession has come a long way since 2015. Indicative of the progress and current developmental needs, the United Kingdom’s regulator of accountants and auditors, the Financial Reporting Council (FRC), on March 30, 2026, published Generative and Agentic AI Guidance - Risks, mitigations and illustrative examples. While noting the benefits and potential of AI, the FRC also stated that “these technologies pose risks to audit quality too, relating to the risk of deficient outputs, the risk that outputs are misused and the risk that audit methodology is not compliant with audit standards.” The guide goes on to describe the characteristics of the risks and mitigating procedures, followed by examples.
The risk of deficient outputs includes:
1. Hallucinations
2. Omissions
3. Distortions
4. Faulty reasoning
5. Inconsistencies
These outputs can be the result of system performance issues and/or input issues. The former relates to how well the system has been developed and tested, while the latter considers the competence and care taken by the user of the system.
The risk that outputs are misused focuses on two areas:
1. Misinterpretation of the output
2. Misunderstanding of the methodology
The risk that audit methodology is not compliant with audit standards enters the area of judgement. The auditor must have a thorough knowledge of the AI procedures and results to determine that conclusions are warranted and just as convincing as traditional audit procedures.
The Guide then presents various ways to mitigate the risks of deficient output, which include the following:
1. Designing and developing the system in a way that is responsive to the intended use;
2. Putting the tool through a robust certification process;
3. Equipping those who use the tool with the appropriate knowledge and implementing business rules to govern that us; and
4. Having a human review and/or oversee the outputs and operations of the tool.
Finally, useful examples are given, applying the principles to 1) the summarization of board minutes and 2) contract review.
Further details can be found at Generative and Agentic AI Guidance - Risks, mitigations and illustrative examples.
