AI · 20 November 2026
KPMG AI agents complete first full audit in parallel test
KPMG ran more than 100 AI agents alongside a traditional audit of an anonymous FTSE 250 company's annual report, reportedly asking sharper questions but missing one regulatory link.
What happened
KPMG has completed a full audit using AI agents for the first time, according to reporting by the Financial Times, with Computerworld also covering the test. The project used more than 100 AI agents to review the annual report of an anonymous British public company listed on the FTSE 250 index.
The AI audit ran in parallel with a traditional audit and was not used as the basis for KPMG's formal approval of the company's financial statements. Within the test, the agents independently gathered and analysed information, documented their findings and formed preliminary conclusions. They also worked together and suggested next steps for the audit.
According to KPMG, the system in several instances examined the material far more thoroughly than its human colleagues and asked better questions. The technology was not flawless, however. On one occasion, a human auditor had to intervene when an agent failed to connect the correct regulations to the company's documents.
Why it matters
For professional services, this moves AI beyond drafting and summarising. Here, agents handled multi-step analytical work, recorded their reasoning and proposed what should happen next. That points to a change in how audit teams could be structured, with people spending more of their time reviewing, challenging and deciding, and less on assembling evidence.
The single intervention is just as instructive. The failure was about context and grounding: the agent did not link the right regulatory requirements to the right documents. In any judgement-heavy service, the quality of the handoff between machine and human will determine whether the speed gains hold up under scrutiny.
By the numbers
- More than 100 AI agents deployed to review a single annual report
- FTSE 250 is the index on which the anonymised British company is listed
- One occasion on which a human auditor had to intervene
The Renascence take
Most coverage will focus on the claim that the AI asked better questions. The more useful finding is the one that went wrong. The agent that missed the regulatory link shows that questioning ability and grounding in the right context are separate capabilities, and only the second one reliably protects the client.
The headline rewards the machine for curiosity, but the design lesson lies in the failure: what matters is who notices when the context breaks. Customer-obsessed operators should therefore stop asking whether AI can replace a step and start mapping where a human must be able to see, challenge and override the output. Build the escalation path before you scale the agents, and measure how often reviewers catch errors, not just how fast the work moves.
Sources
This briefing was written by our Newsdesk, synthesising reporting from the outlets below. Follow the links for the original coverage.
FAQ
Questions we get on this topic
More in AI
Stay ahead of CX
Get the signal, not the noise.
The stories shaping customer experience — plus the Journal and Experience Loom — in your inbox.


