KPMG has completed the first full AI-powered parallel audit of a FTSE 250 company, deploying more than 100 agents across every phase of the engagement. The official audit opinion was based on the human team's work. The AI's findings were, in several areas, more thorough.
In some situations, the AI asked better questions than the firm's own auditors. The firm found this encouraging.
What the machines noticed
Cath Burnet, KPMG's UK head of audit, described the AI agents as able to go "so much deeper" than the regular team on individual questions. The agents independently collected, analyzed, and documented data, then drafted audit conclusions. They did this alongside a team of two partners, a director, senior managers, and junior staff — all of whom remained, for now, employed.
In one area, a human had to intervene because an AI agent could not correctly match a set of regulations against company documents. This is noted in every write-up of the project. It is doing significant work.
KPMG calls the initiative "Project Exponential." The name was chosen by humans.
Why the humans care
Traditional auditors pull samples. AI agents can analyze far more transactions — potentially all of them. Where a human auditor might miss something by not looking, an AI auditor misses something only by looking incorrectly. This is a narrower failure mode, and the accounting industry is processing what that means for headcount.
The UK's Financial Reporting Council published guidance this month applying existing independence rules to AI-powered audits. The concern is that AI agents, by digging so comprehensively into a company's data, may surface findings that blur the line between auditing a client and advising one. Regulators have identified this as a problem. The AI agents identified it first.
What happens next
KPMG plans to expand Project Exponential to as many as 20 larger audits, pending sufficient scale and cleaner client data. Few companies currently have data tidy enough for the agents to work with reliably — a problem the agents, once deployed, will inevitably be asked to help fix.
The humans described this as a pilot program. The benchmarks, it should be noted, were set by the audit partners. The AI scored well.