The accounting and consulting firm removed its October 2025 report on agentic AI, AI systems designed to act autonomously, after UBS, the UK's National Health Service, Swiss Federal Railways, and Transport for London said the report misrepresented their AI work.
KPMG removed a report on agentic AI from its websites after four organizations it profiled publicly said the report misrepresented their work. The October 2025 publication, "Redefining excellence in the age of agentic AI," attributed AI deployments to UBS, the UK's National Health Service, Swiss Federal Railways, and Transport for London. All four told the Financial Times the claims were untrue or misleading, according to TechCrunch's Anthony Ha, citing the FT.
The retraction is the second in a month for one of the Big Four accounting and consulting firms. EY withdrew a report on loyalty rewards programs last month after apparent AI-generated inaccuracies, including footnotes that did not check out, surfaced publicly. The recurrence sharpens the question. If professional services firms are selling AI strategy to clients, what does their own internal verification of AI-assisted research actually look like.
KPMG has not confirmed AI was used to draft the report. The "hallucination" framing comes from GPTZero, a commercial AI-detection vendor whose product depends on flagging AI-generated text. Its diagnosis is best read as labelled expert commentary, not an independent adjudicated finding, until KPMG's own investigation concludes. What is established is the pull, the four on-record contradictions, and the firm's own stated standard.
That standard, from a KPMG spokesperson, is the spine of the incident: "We expect all our people to follow our guidelines on the responsible use of AI, including human oversight to validate content and verify independent sources." The failure is measured against that line, not against an abstract worry about AI reliability. A verification step designed to catch fabricated deployments either did not exist on this report or was not followed.
The four named organizations are concrete. UBS told the FT its characterization in the report was wrong. The UK's National Health Service said the same. Swiss Federal Railways and Transport for London joined them. These are not anonymous industry complaints. They are named subjects of a public-facing advisory document telling the firm, on the record, that the document was not accurate about them.
The open question is procedural. KPMG's statement describes a control: human oversight to validate content and verify independent sources. The agentic AI report suggests that control did not operate, or did not exist in a form that would have caught false attributions about named organizations before publication. The EY precedent suggests this is not an isolated lapse.
What does "responsible AI" verification look like inside a firm that sells AI strategy for a living? KPMG's own guidelines name the step. The next test is whether the firm's broader AI advisory practice is built to perform it, or whether the agentic AI report was treated as a marketing artifact on a faster track than the controls the firm says it follows.