Imagine hiring an assistant who reads every document in the building overnight, then sending the whole team on a course called ‘Don’t Believe The Assistant’. That’s somewhat where the Big Four have landed with AI agents on audits.
KPMG reports putting more than 100 AI agents to work on a live audit of an unnamed UK-listed company in a trial named Project Exponential, with plans to expand the approach to up to 20 larger audits. Running the experiment live is no small gamble, given that real client books and an official audit opinion are on the line.
Meanwhile, EY and Deloitte have both put AI agents into auditors’ hands this year, while PwC US is focusing on training more than 85,000 employees to interrogate AI outputs rather than simply nod along. Across the board, the official line stays identical: algorithms do the legwork, but humans still carry the legal liability.
Three Rollouts In Seven Months
EY kicked things off in April 2026 by rolling out what it calls enterprise-scale agentic AI across its Assurance division. Developed using Microsoft technology and baked into its global audit platform, the agents plug into a system that reportedly goes through more than 1.4 trillion lines of journal-entry data annually.
Not to be outdone, Deloitte arrived on 24 June with a linked network of agents inside its Omnia platform. The digital crew handles early-stage chores like scraping data, sifting through evidence and writing first-draft paperwork, routing every tentative takeaway to a human professional for a reality check.
Come November, KPMG raised the stakes on scale, pairing more than 100 agents with a team of humans on a live audit. While the bots rounded up data and drafted findings, the human auditors called the shots and picked apart the output before anything close to a formal opinion was finalised.
More from Artificial Intelligence
- Why Is Open-Source AI Becoming One Of The Biggest Conversations In Tech?
- Can Europe Build Its Own Speech AI? Reson8 Is Giving Its Answer Away
- Has 2026 Been The Year We Finally Fell Out Of Love With AI?
- The Data Centre Dilemma: Can We Really Complain About Data Centres While Simultaneously Demanding More Of The Technology They Power?
- Why Are AI Companies So Interested In Giving Their Models Memory?
- What Are Small Language Models And How Are They Different From Large Language Models?
- Disney Is Teaching AI Its Creative Style – What Does This Mean For The Future Of Entertainment?
- Your Patient Records Could Become AI Training Data – What Do Privacy Rules Say About De-Identified Health Data?
Did The Agents Stumble?
The agents aren’t flawless, and KPMG’s trial showed it. One agent couldn’t match regulations to the documents it was checking, which is why people stay in charge of the result.
Interestingly, KPMG found that the bots occasionally asked more probing questions and dug deeper into tricky financial details than the humans in the room. Before anyone declares the junior auditor obsolete, though, the technology is still miles away from beating people across the board. The pilot ran alongside a traditional audit, with reviewers checking every conclusion.
Accountability is the other limit. The companies say they aren’t trying to have AI issue audit opinions. The UK’s Financial Reporting Council has warned that sharing AI-derived insights with audit clients can threaten an auditor’s objectivity and independence.
Trust, But Verify (Then Verify Again)
PwC’s answer to the risk is basically a masterclass in raising an eyebrow.
The firm is rolling out a Critical Thinking and Judgment certificate programme to more than 85,000 US partners and employees. It starts in December with a target of June 2027, and it’s built to sharpen their ability to challenge what AI tells them.
The worry behind it is over-reliance, and it’s easy to see why. AI is fast, fluent and extremely sure of itself, even when its output is incomplete or flat-out wrong. An auditor who takes a confident answer at face value without enough evidence is quietly wearing down the scepticism an opinion depends on.
Junior work is changing too, with less mechanical testing and more reviewing and interpreting of whatever the agents produce. It’s much like supervising a keen intern who never gets tired.
Agents Draft, Humans Sign
The Big Four are working to a simple rule for now: the agent drafts and the human signs. KPMG’s plan to move to as many as 20 larger audits will test how far that rule goes when the workload gets heavier.
For all the talk of automation, there’s no evidence yet that these digital assistants are cutting client bills or thinning out audit teams. In fact, the loud emphasis on human oversight hints that the industry still doesn’t trust algorithms to master audit evidence, independence and healthy scepticism on their own. The next round of audits will show how quickly that doubt starts to melt away.
