AI in Auditing: What KPMG’s Agent Trial Means for Australia

Close-up of a black calculator beside a ballpoint pen and a paper clip resting on a printed bar chart.

AI in auditing moved a step closer this week, after KPMG revealed it ran more than 100 AI agents through a full trial audit of a UK-listed company. Headlines have run with the line that the AI “asked better questions” than KPMG’s own accountants. The fine print is narrower: it was a parallel experiment that KPMG did not rely on, and people set the goals, checked the work and stepped in when an agent got stuck.

Here’s what KPMG actually did, what regulators in the UK and Australia say about AI in audits, and what it means for Australian accounting practices and the businesses they audit.

What KPMG’s AI audit trial involved

According to the Financial Times, republished by The Irish Times on 18 November 2026, KPMG deployed more than 100 AI agents this year to review the financial statements of an unnamed FTSE 250 company. The project, code-named Project Exponential, ran alongside a conventional audit carried out by KPMG’s accountants.

  • What the agents did. Agents working on different parts of the audit gathered data, analysed and documented it, and formed draft conclusions and observations. They worked with each other and could suggest next steps.
  • Who built it. A team built the agents in four months. KPMG said two partners and one director worked with senior managers and junior staff to build and run them.
  • Who signed off. An audit partner signed off the experimental audit, but KPMG did not rely on it when it formally signed off on the company’s accounts.
  • What comes next. KPMG plans to expand the project to up to 20 larger audits and hopes to eventually offer AI-based audits to clients.

Where the AI agents did well, and where a person stepped in

KPMG’s UK head of audit, Cath Burnet, told the FT that in several cases “the AI was able to go so much deeper” than the regular audit team, and that in some situations it asked “better questions”. That is a claim about some parts of one trial, not a finding that AI outperforms auditors in general.

In one area of the audit, a person had to intervene after an AI agent failed to match a set of regulations to the company’s documents. Staff also set the agents’ objectives, supplied background on the business and previous audits, checked the agents’ work periodically and challenged their conclusions before the opinion was finalised. Burnet described human involvement as “critical”.

KPMG said its system recorded the agents’ interactions with documents down to individual characters and timestamps, giving what it called “full visibility and full traceability”. That kind of record is what lets a person check and explain an AI-assisted conclusion afterwards.

The fine print: a trial, not AI replacing auditors

It would be wrong to read this as “AI audits are here”. The FT report is clear on several limits:

  • The AI audit was a parallel experiment. The real audit opinion came from the conventional audit.
  • KPMG said the technology isn’t yet available at the scale needed to run thousands of audits.
  • Few companies currently have data “clean” enough for AI agents to work with, according to Burnet.
  • Some companies may prefer not to use an AI-based audit at all.

PwC has also run several “parallel audit” projects to test AI without relying on it, the FT reported. Testing in parallel before trusting the output is becoming the standard way large firms are approaching AI in auditing.

How regulators view AI in auditing

In the UK, the Financial Reporting Council published guidance on auditor independence when using technology, including AI on 16 November 2026. It says technology used appropriately can improve audit quality, but that passing AI-derived insights to an audited company can threaten integrity, objectivity or independence in some circumstances. Quoted in the FT report, the FRC’s Mark Babington put it plainly: “Whatever technology a firm uses, the auditor remains accountable.”

In Australia, the Auditing and Assurance Standards Board (AUASB) issued Impact of AI on Auditors in July 2025. It is an information sheet rather than a new standard, but it sets out practical points for firms using AI audit tools:

  • Tools should be reviewed and approved at firm level, and approved again for each engagement.
  • Staff need training to understand a tool’s limits and avoid “automation bias”.
  • The tool should produce an audit trail showing the data used and the logic applied.
  • Firms should allow time for alternative procedures in case a tool doesn’t work, and consider running it in parallel with other procedures in the first year.
  • Data privacy and security need managing, because much of the data belongs to the audited entity.

That last point about running in parallel is exactly the approach KPMG and PwC have taken.

What AI in auditing means for Australian accounting firms

Most Australian practices won’t be building 100 agents. The useful lesson is the pattern KPMG followed, which applies at any size:

  • Split the work into clear pieces. KPMG pointed agents at different parts of the audit. Smaller firms can start with one well-defined task, such as matching invoices to bank lines or drafting a first pass of working-paper notes. Our post on single agent vs multi-agent AI explains why one well-built agent is usually the better place to start.
  • Keep a person accountable for every conclusion. The FRC and AUASB both put responsibility on the auditor, not the tool.
  • Insist on traceability. If a tool can’t show which document and which figure it relied on, it can’t support audit evidence.
  • Test before you rely on it. Run the tool alongside your normal process on a past file and compare the results.

The same approach suits non-audit work in a practice, from bookkeeping reviews to client queries. Our free AI for accountants course covers practical uses across the whole practice, and the Xero AI course focuses on AI features inside Xero.

What it means for businesses that get audited

If you run a company that is audited, the biggest takeaway is about your records. KPMG’s comment that few companies have clean enough data is a reminder that AI can only check what it can read and match. Consistent document naming, complete supporting records and clean ledgers make any audit easier, with or without AI.

It also helps to watch for fake or altered documents, which AI-assisted checks are increasingly used to spot. We covered that in our post on AI document fraud detection.

Will AI audits change audit fees?

Nobody can say yet. KPMG acknowledged “clear time-saving opportunities” from AI-based audits but declined to say whether this would reduce fees. Audit firms have traditionally billed for large teams of junior staff, so pricing models are likely to be debated as AI tools mature. For now, any fee changes are speculation, and we’d treat confident predictions with caution.

Practical first steps for a smaller practice

  1. List the repetitive checking tasks your team does every month, and pick one with clear inputs and a clear right answer.
  2. Check where the data lives and how clean it is before choosing a tool.
  3. Run any AI tool on a past, completed file in parallel and compare its results with what your team found.
  4. Keep a person approving every output, and keep a log of what the tool did.
  5. Write down your firm’s rules for client data, including where it is stored and which tools may see it.

If a task needs an agent that can work across your own systems, we covered approvals and logging in our post on AI agent safety in Australia.

Caveats

  • Second-hand reporting. The KPMG details come from the FT’s report, republished by The Irish Times. KPMG has not named the company or published the trial’s results.
  • A UK trial. The audit was of a UK-listed company under UK rules. Australian audits follow AUASB standards, and ASIC oversees auditors here.
  • Early days. This was one experimental audit. KPMG’s planned expansion to up to 20 audits will say more about how well the approach holds up.
  • No guaranteed savings. Whether AI saves time or money in a practice depends on the task, the data and how carefully the tool is tested and supervised.

This is general information based on the Financial Times report republished by The Irish Times, the FRC’s 16 November 2026 announcement and the AUASB’s Impact of AI on Auditors information sheet, as at 19 November 2026. Guidance and products in this area change often. It is not accounting, audit or legal advice for your specific situation.

Thinking about AI for your practice?

Start with our free AI for accountants course or the AI agents course to see where AI fits. Our AI business audit (a review of where AI can help your operations, not a financial audit) shows which tasks are worth automating first. Aideveloper builds AI agents and finance AI tools for Australian businesses, with human approval, audit trails and testing built in. Talk to us about your practice.

Featured image: Photo by Pixabay on Pexels.