KPMG has tested more than 100 artificial intelligence agents during an experimental audit of a FTSE 250 company, finding that the systems investigated some financial reporting issues more extensively than its conventional audit team.
The accounting business ran the experiment alongside the statutory audit of an unnamed UK-listed company, using AI agents to examine financial information and carry out connected audit procedures.
KPMG reported that the automated systems identified questions warranting further investigation, although human intervention was required when an agent failed to match regulatory requirements correctly with company documentation.
The experimental work was reviewed by an audit partner, but its findings were not relied upon when the business formally signed the company’s accounts.
The pilot therefore did not replace the statutory audit or transfer responsibility for the final audit opinion to an automated system.
The work was carried out using agents developed over approximately four months, with the technology tested against activities undertaken by the human audit team.
The findings were disclosed as professional services businesses expand their use of AI across financial analysis, audit testing, document examination, and other established processes.
Comparing automated and conventional procedures —
Financial audits require accountants to obtain sufficient appropriate evidence about financial statements, assess material risks, and evaluate whether reported information complies with applicable accounting standards.
The work often involves examining contracts, financial records, transactions, accounting policies, and the assumptions used by company management.
AI agents can perform connected tasks involving the retrieval and analysis of information, allowing large volumes of material to be examined without each individual step being performed manually.
In KPMG’s experiment, the agents identified lines of enquiry that the business considered useful alongside the findings of its conventional audit.
Cath Burnet, KPMG’s UK head of audit, described instances in which the automated work examined particular matters more deeply than the accountants conducting the normal audit.
However, the need for human intervention when an agent incorrectly matched regulations with supporting documents demonstrated a limitation of allowing automated systems to interpret technical requirements independently.
A system may process extensive quantities of information while still reaching an inappropriate conclusion if it applies the wrong rule or misunderstands the relevance of a particular document.
Audit procedures therefore require more than the identification of apparent anomalies. The evidence must be assessed for reliability, relevance, and its significance to the financial statements.
The pilot also did not establish that AI agents consistently outperform qualified auditors or that an automated audit could meet statutory requirements without professional oversight.
KPMG has previously incorporated AI into its Clara audit platform. The latest experiment extends that work by testing multiple agents across connected audit activities.
Accountability and independence remain central —
The Financial Reporting Council has published guidance on generative and agentic AI, including how audit businesses should evaluate technology risks and maintain appropriate oversight.
Its March 2026 guidance addresses the possibility that auditors may misunderstand how an AI tool fits within their methodology or rely on outputs without sufficient evaluation.
The regulator has emphasised that professional responsibility remains with the auditor, regardless of the technology used to perform particular procedures.
Separate guidance issued in November addresses the ethical risks that can arise when audit businesses share technology-derived insights with their clients.
An audit may identify information that is commercially useful to the company being examined. Providing additional analysis or technology services based on that information could create questions about whether the auditor’s objectivity or independence is compromised.
The FRC’s guidance requires firms to consider those risks in the context of their professional obligations rather than assuming that AI-generated insights can automatically be offered as additional services.
These matters are separate from whether AI can improve the technical quality or efficiency of audit procedures, but both affect how the technology can be used in practice.
For audit committees and company directors, greater automation also increases the importance of understanding which procedures were performed, how outputs were checked, and who approved the conclusions.
KPMG has not established that the experiment will lead to lower audit fees or a reduction in the number of accountants required to complete statutory engagements.
Its parallel audit provides evidence of the analytical potential of AI agents, alongside a documented example of why human interpretation and final professional judgement remain necessary.




