How the Trial Worked

KPMG used more than 100 AI agents to examine the accounts of an unnamed FTSE 250 company this year. The test ran alongside a normal audit by KPMG's own accountants. The project is called Project Exponential. The Financial Times first reported the trial on Wednesday.

The agents worked across the whole audit, not just on single tasks. They collected data, analyzed it, wrote notes, and drafted conclusions. They also worked with one another and suggested next steps. A team built the agents over four months.

What KPMG Found

Cath Burnet, KPMG's UK head of audit, said the AI went "so much deeper" in several cases. She also said it asked better questions in some situations. Burnet added that the AI team had a different shape and size from a traditional audit, but human involvement remained critical.

Humans were still needed. An auditor had to step in when one agent failed to match a set of rules with the company's documents. Staff also set goals for the agents, shared background on the business, and reviewed their work. A partner signed off on the test, but KPMG did not use its findings to approve the company's accounts.

Plans and Limits

KPMG might extend the project to as many as 20 larger audits. In time, it wants to offer AI-based audits to clients. The firm says AI offers clear time savings, but it did not say if that would lower audit fees.

The firm also said the technology is not ready for thousands of audits. Some companies do not have clean enough data for the agents to work well. Rival firm PwC has run similar side-by-side audit projects.

Audit firms have long used large teams of junior accountants for labor-intensive work under senior staff. The trial points to a possible shift in how audits are done, and it raises questions about the future role of accountants.

Regulators and Trust

The UK's Financial Reporting Council (FRC) states that auditors remain accountable, whatever technology they use. They should be able to explain how they used it, challenge its outputs, and stand behind the evidence. Beyond that, the regulator has warned that sharing AI-generated insights with audit clients could put auditor independence at risk.

A report in October found that 23% of audit committee chairs had limited confidence in AI use in audits. Concerns included over-reliance, weaker professional skepticism, data security, and false information. KPMG says its system records every agent action with timestamps, so the work can be traced.