KPMG had the accounts of a FTSE 250-listed company examined by more than 100 artificial intelligence agents, alongside the audit conducted by its accountants. The AI is said to have sometimes asked better questions than the humans, but the accounts were not approved on this basis.
It was the Financial Times that revealed it on November 18, 2026: KPMG conducted its first “AI-based” audit of a UK-listed company this year. The company is not named, but it belongs to the FTSE 250. Its financial statements were scrutinized by more than 100 agents, while the firm’s accountants carried out the traditional audit. The experiment has a codename: Project Exponential.
Should we therefore speak of an audit performed by an AI? The shortcut is tempting, but an audit conducted in parallel is not an audit entrusted to a machine.
KPMG built AI agents for the entire audit process in 4 months
These agents did not just perform isolated tasks. They collected, analyzed, and documented data, then drafted conclusions and provisional observations. Those working on different parts of the audit communicated autonomously with each other and could propose the next steps. A team of two partners, a director, senior managers, and juniors took four months to build them, and KPMG refuses to specify its exact size.
Cath Burnet, who heads KPMG’s audit in the UK, told the Financial Times:
“In several cases the AI was able to go so much deeper”
In other words, the agents dug deeper than the human team on several points, and they sometimes asked better questions than the accountants. She adds that the agents’ interactions with documents are recorded character by character, with timestamps, and that the project team had a different shape and size from that of a traditional audit.
KPMG did not approve the accounts based on the AI agents’ audit
This is the detail that a headline doesn’t always tell: according to the newspaper, KPMG did not rely on the experimental audit to formally approve the company’s accounts. A partner did validate the exercise. The humans remained in control throughout the trial: they set the agents’ objectives and provided them with information about the company and previous audits, then they checked their work at regular intervals and challenged their conclusions before the final opinion.
And the AI did not succeed at everything. At one point, an agent failed to link a set of regulations to the company’s documents, and a human had to intervene. Cath Burnet acknowledges other limitations: the technology is not ready for thousands of audits, and few companies have data clean enough for the agents to work properly. Some might even refuse an AI-conducted audit. KPMG nonetheless wants to extend the trial to a maximum of 20 larger audits and hopes to eventually offer AI-based audits to its clients.
Audit and AI: UK regulator reminds that the auditor remains responsible
The Financial Reporting Council (FRC), the UK’s audit regulator, published guidance on auditor independence when using technology, including AI, on November 16, 2026. It covers the analyses that the auditor draws from these tools and communicates to the management or board of an audited company. These exchanges can, under certain circumstances, threaten their integrity, objectivity, or independence, especially when the work producing them falls outside the scope of the financial statement audit. The FRC explains that it drafted this text at the request of auditors.
Its executive director of regulatory standards, Mark Babington, writes that AI can improve the quality of audit work, but that auditor independence must remain at the heart of the mission. He also told the Financial Times that the auditor remains responsible, regardless of the technology used. Trust, however, is far from assured. According to an October report from the Centre for Public Interest Audit, cited by the newspaper, 23% of audit committee chairs have limited confidence in the use of AI in their company’s audit, and the report notably highlights the risk of excessive dependency and a weakening of critical thinking.
KPMG does not present its job cuts as an effect of AI
The employment question remains. Reports on the trial do not address it directly: audit traditionally relies on large teams of juniors supervised by seniors, without specifying what AI would change for these roles. Recent cuts at KPMG in the UK do not settle the question either. On September 15, the Financial Times reported a plan to cut around 200 jobs in consulting, representing about 4% of the workforce in that business, including AI and cybersecurity teams.
The affected employees heard that the firm needed to “right-size” its capabilities. Six months earlier, nearly 600 jobs had already been put at risk, including over 400 in audit, due to a low attrition rate and sluggish demand. AI is therefore among the affected professions, without appearing in the reasons given, even though the newspaper notes that its rapid adoption raises questions about the future of a very labor-intensive sector.
In February, the Financial Times revealed that KPMG International had secured a fee reduction from its own auditor, Grant Thornton, by arguing that AI would make the work less costly. For its clients, the firm acknowledges clear time savings, but does not say whether fees will drop: it remains to be seen if the argument works both ways.
