KPMG has withdrawn a report on enterprise AI adoption after four of the organizations it cited as case studies said the claims about their AI work were untrue or misleading. The October 2025 report, titled "Redefining excellence in the age of agentic AI," was pulled from KPMG's websites this week while the firm investigates how the inaccuracies got in. The detection firm GPTZero identified the errors and attributed them to AI hallucinations — meaning a professional services firm appears to have used AI to help write a report about AI.
UBS, the UK National Health Service, Swiss Federal Railways, and Transport for London were all named in the report as examples of organizations deploying agentic AI in production. All four told the Financial Times the descriptions of their work were either wrong or significantly overstated. None of the four had been contacted to verify the case studies before publication, according to the reporting that surfaced the dispute.
The episode is a clean illustration of the failure mode auditors and consultants have been warning their own clients about for two years. A large language model, asked to enrich a report with concrete enterprise examples, will produce concrete enterprise examples — whether or not those examples exist. Without a human checking each citation against a primary source, the fabricated detail flows straight into the final PDF. The case studies in the KPMG report read as authoritative precisely because the firm's brand is authoritative.
Key facts
- 01KPMG withdrew 'Redefining excellence in the age of agentic AI,' first published in October 2025, after disputes over its factual claims.
- 02UBS, the UK National Health Service, Swiss Federal Railways, and Transport for London all said the report's claims about their AI usage were untrue or misleading.
- 03Research group GPTZero traced the inaccuracies to AI hallucinations — meaning KPMG appears to have used AI to help write a report about AI.
- 04Last month, EY withdrew a separate report on loyalty rewards programs that contained fake footnotes and AI hallucinations.
GPTZero, which sells AI-detection tools to publishers and universities, flagged the inaccuracies after running the report through its analysis pipeline. The firm told the FT the patterns it identified were consistent with model-generated text that had not been fact-checked against external sources. It is unusual for a Big Four firm to have a third-party detection vendor catch errors of this kind in a flagship publication.
KPMG's response so far has been a takedown and a statement of policy rather than an explanation of what happened. The firm has not said which team produced the report, which model or tool was used, or whether the case studies were drafted by AI from the start or inserted into a human-written document during editing. Until KPMG publishes findings from its internal investigation, the most that can be said is that the firm's own AI-use guidelines were not followed for this report.
“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.”— KPMG spokesperson, KPMG
The KPMG statement reads as boilerplate, but the underlying policy — human oversight, source verification — is the exact control that, if applied, would have caught the errors before publication. The fact that it didn't is the story.
This is the second high-profile retraction by a Big Four firm in as many months. EY withdrew a report on loyalty rewards programs last month after readers spotted fake footnotes and citations to papers that did not exist, also attributed to AI hallucinations. Two such incidents in five weeks suggest that the rollout of generative AI inside professional services has run ahead of the editorial controls those firms apply to client-facing work.
The commercial stakes here are real. KPMG, EY, and their peers sell AI advisory engagements to the same banks, health systems, and transport agencies that turned up as fabricated case studies. A consulting deck recommending an AI deployment pattern is worth less if the firm producing it cannot reliably check whether its own published examples are true. The reputational risk also runs the other way: organizations named in vendor reports without their consent now have to consider whether to monitor third-party publications for false attribution.
There is a defensible version of how AI gets used in research workflows — drafting, summarization, structuring, with every external claim verified against a primary source by a human before publication. That workflow is slower than letting the model write the case studies, which is presumably why it was skipped. The skipped step is the entire value of the audit-firm brand.
For the AI industry, the KPMG and EY retractions are an unhelpful data point at an inopportune moment. Enterprise buyers are currently being asked to trust agentic systems to take actions on their behalf — drafting contracts, filing reports, executing workflows — on the premise that hallucinations are a solved or solvable problem in production deployments. When the firms selling that premise cannot keep hallucinations out of their own marketing collateral, the sales motion gets harder. The fix is not more sophisticated models; it is the human-in-the-loop verification step that KPMG's own guidelines already required, and that someone, somewhere in the production chain, decided to skip.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
