SIAgentNews.com
← All news

KPMG puts AI agents inside its U.S. audits, with auditors still signing off

KPMG is building AI agents into the audit process itself, and people stay accountable for every conclusion. That is a useful template for any high-stakes work.

· First reported by Unite.AI

more than 125U.S. teams KPMG says are getting KCi

The short version

  • KPMG announced its KCi rollout in the U.S. on November 17, built on its existing Clara audit platform
  • Agents help with risk assessment, transaction analysis, coordination and testing support
  • Data pipelines can examine complete transaction populations in selected audit areas
  • KPMG says some engagements ran up to three and a half weeks shorter
  • Next target: all private company audits in 2027; public company audits get a separate, phased plan

Why it matters

A large accounting firm using agents in a regulated field shows the approach can work when controls are in place. The pattern is simple and portable: the agent assembles and flags, and a named person reviews and approves. If your work touches clients, money or regulators, that split is worth copying.

What KPMG is actually rolling out

KPMG is bringing a set of AI agents, meaning software programs that carry out multi-step tasks with some autonomy, into its U.S. audit practice. As Unite.AI reported, the product is called KPMG Clara intelligence, or KCi, and it extends the firm's existing Clara audit platform. The announcement came on November 17.

The agents sit inside the work audit teams already do: assessing risk, analyzing transactions, coordinating the engagement and supporting testing. That differs from handing auditors a separate chatbot and asking them to copy information back and forth between tools.

KPMG says more than 125 U.S. teams and 500 teams worldwide are part of the rollout. The stated next goal is every private company audit in 2027, with public company audits following a separate, phased schedule. Those are plans, not a claim that the system already runs every engagement.

What the agents do, and what they don't

In planning and risk assessment, agents look over engagement data, draft early thoughts on where problems might lie, and prepare first-pass documentation. The hope is that teams see trouble spots sooner and have more time on difficult areas before the year ends. A plausible draft is still only a draft, and a person has to evaluate it.

Other agents handle logistics: tracking milestones and requests for information, bringing up relevant technical accounting guidance, and feeding team dashboards. It is unglamorous work, but it keeps an audit moving and makes clear to a client's accounting staff what is being asked of them.

KPMG also describes automation of parts of control testing (checking that a company's internal safeguards work) and of sample evaluation. Auditors stay responsible for reviewing those results. The effort shifts away from putting material together and toward directing it, supervising it and judging whether the evidence supports the findings.

Looking at every transaction, with caveats

The boldest claim is about coverage. Traditionally, auditors check samples of a company's transactions. KPMG says its automated data pipelines can analyze the full set of transactions in certain audit areas, picking out unusual items, trends and anomalies for closer investigation.

The qualifiers matter. Covering everything within a chosen area is not the same as automating the whole audit. And an item the system flags is not proof of an error. It is a starting point that tells auditors where to look.

KPMG reports that on some engagements this approach trimmed as much as three and a half weeks from the timeline. It also spared clients from manually pulling large numbers of supporting documents. This is a company-reported result from selected jobs, not an average across all of them, so it should not be read as typical.

Keeping humans responsible

KPMG says KCi runs under its audit methodology, professional standards and what it calls its Trusted AI framework. Teams can follow an output back to the data behind it, examine that data, and confirm findings independently before anything enters the firm's official record.

That traceability is the core of the design. Agents can generate and organize material, but accountability stays with people. Christian Peo, a vice chair in KPMG's U.S. audit and assurance practice, described the goal as giving auditors more time to understand information and question what looks wrong, rather than replacing their judgment.

On the technology side, KPMG combines its own tools with enterprise AI available through relationships with Anthropic, Google Cloud, Microsoft and OpenAI. The announcement does not say which provider powers which function, so those names are not a map of how KCi is built.

The people side of the change

KPMG is also reworking jobs around the software. It is redefining roles, scheduling supervision and review earlier in an engagement, and organizing client data in a more uniform way. The firm says nearly 1,000 audit and assurance interns went through training at its Lakehouse facility earlier this year, covering critical thinking, professional skepticism and how to assess work that AI helped produce.

The lesson is that putting agents into a profession built on accountability takes more than buying a model. Clean data, the timing of reviews, traceable outputs and trained staff all determine whether quicker preparation becomes better audit work.

The 2027 expansion will be the real test of whether the model holds up consistently across many engagements while keeping the human scrutiny that audit conclusions depend on.

What to watch

  • Whether KPMG meets its 2027 goal for private company audits
  • How the separate, phased rollout for public company audits is handled
  • Whether results like the three-and-a-half-week saving show up across more engagements

Try it

Next time an AI tool drafts something a client or payment depends on, name one person who must review it and write down what they check before approving.

Sources and links

First reported by Unite.AI. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news