Rudra Kasturi Inc.

Article · AI News

KPMG tests an audit run by more than 100 AI agents: why human auditors still matter

SUMMARY

KPMG has tested more than 100 AI agents alongside a conventional audit of a UK-listed company, but the experiment was not used to formally approve the accounts.

KPMG has completed an experimental audit using more than 100 AI agents to examine the financial statements of an unnamed FTSE 250 company alongside a conventional audit, according to the Financial Times. The trial offers a glimpse of how agent-based systems could change professional services, but it does not mean an AI system has replaced the auditor.

The Financial Times reported that KPMG’s UK audit team used the agents to investigate issues and examine company documents. The firm’s audit head said the agents sometimes explored questions more deeply and raised better questions than human accountants. The same report also describes a failure that required human intervention: one agent did not correctly match regulations to the relevant company documents.

Importantly, the experimental work was not used to formally approve the company’s accounts.

What did the AI agents do?

An AI agent is software that can pursue a defined task through multiple steps, using tools and other agents where configured. In an audit, a group of agents could be assigned separate jobs such as checking figures, finding unusual transactions, matching documents and identifying areas that need further investigation.

The attraction is scale. An agent can review large volumes of information and flag patterns for an auditor to examine. But finding a pattern is not the same as deciding whether it is material, whether evidence is reliable or whether a financial statement complies with the rules.

Why human oversight remains central

The reported document-matching failure illustrates a key risk. An agent may produce a plausible answer while applying the wrong regulation or drawing a conclusion from incomplete evidence. Audit work also involves professional scepticism, accountability and judgement.

KPMG’s approach, as described by the Financial Times, kept people involved in setting objectives, checking outputs and challenging conclusions. The firm reportedly plans to expand the work to as many as 20 larger audits, while acknowledging that the technology is not ready for thousands of audits and that some company data is unsuitable.

What businesses should learn

  1. Use agents first for evidence gathering and repetitive checks, not final sign-off.
  2. Keep a traceable record of the documents, instructions and steps behind each result.
  3. Require a named human owner for every material conclusion.
  4. Test the system against known errors and incomplete records.
  5. Measure time saved alongside false positives, missed issues and review effort.

The real test for AI in professional services

The question is not whether 100 agents can perform 100 tasks. It is whether their work improves the quality and speed of the final audit without weakening independence, evidence standards or accountability.

Bottom line: KPMG’s trial suggests AI agents may become useful audit assistants. The experiment also shows why responsibility for the audit opinion must remain with qualified human professionals.

Sources