An investigation by AI detection firm GPTZero, independently verified by the Financial Times, found that KPMG's October 2025 report on autonomous AI systems held together with hallucinated footnotes.
A KPMG-branded report meant to show companies how to use the next generation of AI was treated as a piece of authoritative research, then quietly cited by the very audiences who should have been checking it. "Big Four" is industry shorthand for the four largest global accounting and consulting firms — KPMG, Deloitte, PwC, and EY — and KPMG sits in that group. The report, "Total Experience: Redefining Excellence in the Age of Agentic AI," was published in October 2025 as a guide to "agentic AI," the term for autonomous systems designed to take multi-step actions on a user's behalf rather than simply answer questions. It contained 45 footnotes. Only 5 of them pointed to real sources, according to an investigation by AI-detection firm GPTZero, independently verified by the Financial Times.
The other 40 citations did not survive a closer look. GPTZero's count put 28 in a middle category that paraphrased the titles of real sources but added fake components, and 12 as too vague to verify at all. Several entries described AI products that, on inspection, did not exist or did not have the capabilities the report attributed to them. GPTZero called the pattern "vibe citing" — plausible-looking references that hold up on the page but collapse the moment anyone opens the link.
The KPMG report's named examples show the pattern in human-scale form. It credited Emirates' "Sara" mobile assistant with capabilities — including the power to alter flights — that Emirates says it does not have. (Sara, launched in 2023, is a conventional mobile assistant, not an AI-powered chatbot.) It described agentic AI deployments at UBS across "investment advisory, risk management and compliance monitoring" that UBS told the Financial Times were "factually incorrect." It claimed Swiss Federal Railways (SBB) had AI agents that could help passengers plan, book, and optimize trips based on preferences, real-time conditions, and carbon impact; SBB called that "not accurate."
GPTZero chief executive Edward Tian put the stakes plainly: error-riddled papers published by the Big Four could "poison the well of information" and lead to second-hand AI hallucinations. The investigation is the kind of audit a reader is now expected to run on any institutional report that looks authoritative in the AI era — because a letterhead tells you who paid for the work, not whether the work holds up.
KPMG has since pulled the paper and is "reviewing the circumstances surrounding its publication," a KPMG spokesperson told the Times. The spokesperson said the firm "takes the accuracy and integrity of its published content seriously." As of Engadget's June 13, 2026 report, no further public correction had been issued.
The structural lesson sits below the embarrassment. Big Four research products move through the economy the way wire-service copy once did: into consulting decks, into enterprise procurement briefings, into academic literature reviews, and into the next round of institutional reports. When a flagship study on agentic AI is hollowed out by AI-generated references, the failure does not stay on KPMG's letterhead. It propagates.
A reader who treats the report's findings as a guide to buying agentic AI is not buying KPMG's analysis. They are buying a hallucination vector that survived enough internal review to carry the firm's name. The next move is to read the footnotes.