← The feed Many Minded · The brief
KNOWLEDGE · forward · impact 2/5 · 2026-11-18 · KPMG

KPMG ran 100+ AI agents through a shadow audit of a FTSE 250 firm

In a parallel test, KPMG's agents sometimes probed deeper than the human team. No cost or quality figures yet, and the official opinion ignored them.

KPMG has run an experimental audit of a FTSE 250 company in which more than 100 AI agents worked across every phase of the annual audit. The firm calls it Project Exponential, and the Financial Times reported it. The agents collected, analyzed and documented data and drafted audit conclusions. The official audit opinion still rested on the conventional human audit, not the agents' output.

Cath Burnet, KPMG's UK head of audit, said the AI went much deeper than the regular team in several cases and sometimes asked better questions. It was not flawless: a human had to step in on one area because an agent could not correctly match a set of regulations against company documents. Two partners and a director worked with senior managers and junior staff, setting goals, giving context, reviewing work periodically and challenging conclusions. KPMG says the system logs agent interactions with documents down to individual characters and timestamps.

The abundance angle is knowledge assurance. Audits are a costly gate on trusted financial information, and agents that probe more thoroughly could make that checking cheaper and broader over time. There are early signs of pressure on pricing: KPMG International reportedly negotiated lower fees from its own auditor, arguing AI makes the work cheaper. KPMG would not say whether client fees will fall.

Watch the scale-up. KPMG plans to extend the project to as many as 20 larger audits, but Burnet said the technology isn't yet ready for thousands of audits and few companies have data clean enough for reliable agent use. No results on cost, time or quality have been published, and the findings come from KPMG's own executive. The UK regulator FRC has also issued guidance applying independence rules to AI, warning that sharing findings beyond the audit, such as cost savings, could look like a sales pitch.

Source: The Decoder