KPMG has run a full audit of a UK-listed company with more than 100 AI agents working in parallel to its human team, and its UK head of audit says the software sometimes investigated financial issues more deeply than her own accountants. The experiment, reported by the Financial Times on Wednesday and confirmed in interviews with KPMG, covered an unnamed FTSE 250 company and ran alongside a conventional audit rather than replacing it.
The project was codenamed Project Exponential. According to the FT account, a team spent four months building the agents, which were then pointed at every stage of the engagement rather than at isolated tasks. The agents collected data, analysed and documented it, drafted conclusions and observations, handed work to each other autonomously, and proposed the next steps. Staff set the objectives, supplied context about the business and prior audits, reviewed the output and challenged conclusions before anything reached the file.
"In several cases the AI was able to go so much deeper," Cath Burnet, KPMG UK head of audit, told the FT, adding that the technology had asked "better questions" in some situations. The team also had a different shape and size from a normal engagement, with two partners, a director and senior managers supervising agents that junior staff would ordinarily have supported.
It was not a clean sweep. Burnet described an auditor having to step in after an agent failed to match a set of regulations to the company documents, a reminder that retrieval failures look like judgement failures until a human checks them. An audit partner signed off on the experimental exercise, but KPMG did not rely on the findings when formally approving the accounts, and humans still set objectives and made the final call.
KPMG now plans to extend the approach to as many as 20 larger audits and eventually to offer AI-based audits to clients. Burnet was careful about the limits: the technology is not yet available at the scale needed to run thousands of audits, and some companies do not hold data clean enough for agents to work effectively. The firm also said the work offers "clear time-saving opportunities" without saying whether audit fees would fall.
That reticence matters because audit economics rest on a pyramid of junior accountants doing labour-intensive testing under partner supervision. Agents that draft conclusions attack the bottom of that pyramid directly. Rival PwC has run its own parallel audits, with global assurance leader Andy Hammond telling the FT the point was to see how AI performs across the whole process without relying on it.
The oversight question is now explicit. The UK Financial Reporting Council published guidance on 16 November on maintaining auditor independence when using technology, and its executive director of regulatory standards, Mark Babington, said that whatever technology a firm uses, the auditor remains accountable and must be able to explain how it was used. A Centre for Public Interest Audit report in October found that 23 percent of audit committee chairs had limited confidence in AI use on their financial statements. KPMG says its system records agent interactions with documents down to individual characters and timestamps, a firm claim that has not been independently verified.
Audit is turning into the first white-collar profession where agents have moved from drafting to concluding, and the binding constraint is not model capability. It is whether a firm can produce an evidence trail a regulator will accept, and whether a partner is willing to put a signature on work a machine assembled. KPMG has shown the workflow can run. The pricing and the liability have not been settled, and those decide whether the 20-audit expansion becomes an operating model or a demonstration.
Comments (0)
Log in to join the discussion
Log InNo comments yet