There is a particular irony – simultaneously instructive and embarrassing – in the fact that a major professional services firm published a report celebrating the transformative potential of agentic artificial intelligence only to have that report withdrawn because it appears to have been substantially produced by the very technology it was praising. KPMG has pulled its October 2025 publication titled “Redefining Excellence in the Age of Agentic AI” after multiple organizations named in the document stated that the claims made about their AI deployments were either entirely false or materially misleading. NEWSCENTRAL believes this incident warrants treatment not as an isolated embarrassment but as a diagnostic signal about the current state of AI-assisted professional publishing.
The mechanics of the failure are instructive. Investigators reviewing the 45 citations contained in the report found that only five accurately pointed to real and verifiable source material. The remainder ranged from subtly distorted to partially fabricated to so vague as to be unverifiable – a phenomenon that has been characterized as “vibe citing,” wherein an AI system generates plausible-looking references that disintegrate under scrutiny. Among the organizations that publicly challenged their characterizations were UBS, which was described as integrating AI agents across investment advisory, risk management, and compliance monitoring through a composable platform co-developed with a major cloud provider – a claim UBS rejected outright. The UK’s National Health Service, Swiss Federal Railways, and Transport for London similarly disputed the report’s representations of their technology deployments.
“This is a governance failure as much as a technology failure,” argues Freddy Miller, Senior Analyst at NEWSCENTRAL. “The hallucination problem in large language models is well-documented. What is new here is that a firm of KPMG’s scale appears to have deployed these tools in a context requiring citation-level factual precision without implementing the verification layers that any responsible workflow demands. The result is that KPMG has now inadvertently provided one of the most compelling arguments against unsupervised AI content generation.”
The institutional response has been measured but telling. A KPMG spokesperson confirmed the report’s removal and stated that the firm expects all personnel to adhere to guidelines on responsible AI use, including human oversight and independent source verification – an acknowledgment that those standards were not met in this instance. The incident follows a comparable episode in which EY withdrew a report on loyalty rewards programs after it was found to contain fabricated footnotes and hallucinated content, suggesting that the problem is not idiosyncratic to one firm but reflects a broader pattern of insufficient editorial governance across the consulting sector as AI tools have been rapidly adopted.
“The systemic risk here is reputational contagion,” cautions Nathan Clark, Enterprise IT and Systems Architecture Analyst. “Enterprise clients rely on Big Four research to inform procurement decisions, regulatory posture, and technology strategy. If the integrity of that research cannot be assured because AI generation is proceeding without adequate human validation, the entire category of consulting-produced intelligence becomes suspect.”
What NEWS CENTRAL finds most significant here is not the error itself – language models hallucinate, and this is a known and documented limitation – but the organizational conditions that allowed the output to travel from initial generation to published distribution without correction. The incident establishes a clear benchmark for what constitutes negligent AI deployment in a professional context, and it arrives at a moment when regulators in multiple jurisdictions are actively developing frameworks for AI-generated content accountability. Enterprises that have not yet formalized AI oversight protocols should treat this episode as a precise illustration of the consequences of delay.