KPMG tests 100+ AI agents in live audit as FRC flags independence risk
KPMG deployed more than 100 AI agents alongside the regular audit of a UK-listed company in a project it calls Project Exponential, according to The Decoder. Cath Burnet, KPMG’s UK head of audit, said the agents sometimes asked “better questions” than the human audit team, a claim KPMG has not backed with independent metrics. The UK Financial Reporting Council (FRC) warned that because agents can analyze far more data than traditional sampling, they may surface findings that could be used as a sales pitch for consulting work and threaten auditor independence.
Verified and unconfirmed
- Verified: The Decoder reported that KPMG ran more than 100 AI agents alongside a live audit of a UK-listed company, and that KPMG’s UK head of audit Cath Burnet said the agents sometimes asked better questions than the audit team; The Decoder also reported the FRC’s independence warning.
- Not yet confirmed: The client, the agent models or vendor, the audit’s duration and cost, and any quantitative comparison between agent and human findings were not disclosed; the “better questions” claim rests on KPMG’s account rather than an independent evaluation, and The Decoder does not quote the FRC directly.
Background
Financial audits have traditionally relied on sampling: auditors test a subset of transactions and rely on management to explain anomalies, because examining every entry is impractical. Auditor-independence rules — in the UK set out by the Financial Reporting Council (FRC), the accounting regulator — restrict audit firms from selling consulting services to the companies they audit, so that issues found during an audit do not turn into a sales pipeline. KPMG’s exercise, which it calls Project Exponential, tests what happens when AI agents can scan far more of a company’s data than a sample would cover, which is exactly the boundary those independence rules are meant to police.
Why it matters
For audit firms, the experiment points to a governance trade-off: agents that review far more data than traditional sampling can generate additional evidence, but firms will need clear rules separating audit findings from consulting opportunities to address the FRC’s independence concern. Audit teams should expect scrutiny of how agent outputs are documented and who is allowed to act on them.
Comments
Comments use a GitHub account and are moderated. Comment rules