Editorial illustration for KPMG's AI agents dug deeper in audits, raising client independence questions
KPMG AI Agents Audit FTSE 250 Firm, Raising Ethics Questions
KPMG's AI agents dug deeper in audits, raising client independence questions
KPMG ran more than 100 AI agents through a full parallel audit of a FTSE 250 company this year, the first time the firm has set an automated system loose on every phase of a real UK-listed audit. The experiment, dubbed "Project Exponential" internally, ran alongside the conventional audit team rather than replacing it. The official audit opinion still came from the human-led process, but the AI version gave KPMG a side-by-side comparison of how machines handle the same work as its own staff.
The results, first reported by the Financial Times, surprised people inside the firm. The agents didn't just match the human team's output. On certain questions, they pushed further than the auditors assigned to the same client, pulling in more documentation and drafting fuller conclusions on their own.
That raised an obvious question for a profession built on independence and judgment: what happens when the software starts outperforming the people it's meant to assist. KPMG also found at least one clear limit, a case where an agent misapplied regulatory text against company records and needed a person to correct it. Cath Burnet, the firm's UK head of audit, has been among those describing what the project showed.
KPMG deployed more than 100 AI agents across all phases of an experimental parallel audit of a UK-listed company for the first time. According to KPMG's UK head of audit, the agents dug deeper into individual questions than the regular audit team. A human had to step in when an AI agent tried to match regulations against company documents.
Why this matters
KPMG's experiment is a preview of a problem every regulated profession will hit once agents start doing real analytical work: the tools are good enough to find things a human reviewer might miss, but nobody has settled what you're allowed to do with what they find. The detail about an agent needing human correction while matching regulations to documents is the tell. These systems aren't just faster versions of checklists, they're making judgment calls that used to require a licensed professional's sign-off, and audit independence rules were written long before that was possible.
For founders building in regulated spaces, the lesson isn't "AI finds more," it's that deployment now runs into decades-old compliance architecture that assumes a human drew every conclusion. KPMG running over 100 agents on one UK audit shows the industry is already past pilot-stage caution. The actual gating factor isn't technical capability anymore, it's whether firms can share AI-generated findings with clients without triggering independence violations.
Watch for how regulators answer that, because it will decide how fast audit, legal, and financial advisory work actually adopt agents, not how capable the agents are.
Common Questions Answered
What was KPMG's Project Exponential and how did it differ from a standard audit?
Project Exponential was KPMG's experimental parallel audit of a FTSE 250 company where more than 100 AI agents were deployed across all audit phases for the first time. Unlike a replacement system, the AI agents ran alongside the conventional human audit team, allowing KPMG to compare how machines handled the same work as their staff while the official audit opinion still came from the human-led process.
Did the AI agents in Project Exponential perform better analysis than KPMG's human audit team?
According to KPMG's UK head of audit, the AI agents dug deeper into individual questions than the regular audit team, suggesting they identified issues that human reviewers might have missed. However, the experiment also revealed limitations, such as when a human had to intervene when an AI agent attempted to match regulations against company documents.
What regulatory and independence concerns does KPMG's AI audit experiment raise?
The experiment highlights a critical problem for regulated professions: while AI tools are capable enough to find things human reviewers might miss, there is no established framework for what auditors are allowed to do with findings generated by AI agents. This creates uncertainty about client independence and audit liability when AI systems make judgment calls that traditionally required human professional expertise.
Why did a human need to step in during the AI agents' regulatory matching process?
When the AI agents attempted to match regulations against company documents, they required human correction and intervention. This demonstrates that despite their analytical capabilities, the AI systems are not yet fully autonomous in making complex professional judgments that involve interpreting regulatory requirements in specific business contexts.
Further Reading
- AI asked ‘better questions' than KPMG accountants in pilot audit of effectiveness - The Irish Times
- Watchdog warns audit firms not to use AI to woo advisory clients - City A.M.
- KPMG finds AI agents ask better questions than human auditors in trial - Business Standard
- KPMG runs first end-to-end AI audit of UK-listed firm: report - The Daily Pioneer
- KPMG’s AI agents audited a FTSE 250 company in a pilot - Sulat