KPMG Fabricated AI Case Studies in a Report Designed to Sell Clients on AI Adoption

The report was titled Redefining excellence in the age of agentic AI. It was designed to sell enterprise clients on AI adoption services. It contained false claims about AI implementations at four named organizations — and none of them recognized the implementations attributed to them.

What the Report Said and What Was Wrong

KPMG’s report cited AI use cases at UBS, the UK’s NHS, Swiss Federal Railways, and Transport for London. All four institutions disputed the claims made about them. GPTZero — an AI detection platform — uncovered the fabrications and verified them against organizational records. The Financial Times confirmed the findings.

The root cause appears to be what GPTZero terms “vibe citing”: citations described as “mostly loose paraphrases of real sources, often missing URLs or correct authors,” with some references having no matching original source whatsoever.

KPMG subsequently withdrew the report from multiple websites.

The Secondary Hallucination Problem

GPTZero’s analysis introduces a useful frame for what this incident creates in practice. When a major consulting firm publishes authoritative-seeming AI case studies that are fabricated, those case studies enter the research ecosystem. They get cited by other reports, scraped by AI training pipelines, and repeated in executive briefings. GPTZero CEO Edward Tian calls these secondary hallucinations — flawed information that spreads from ostensibly credible sources rather than from raw model outputs.

The compounding effect is the distinctive risk here. A hallucination from a generic AI assistant is easily discarded. A hallucination wrapped in a Big Four consulting brand and directed at C-suite audiences is harder to challenge and spreads further before correction.

The Structural Problem

KPMG was marketing AI adoption services while publishing a document that demonstrated inability to manage the research practices the technology requires. The failure is not just an editorial error — it is a demonstration that the governance gap in enterprise AI consulting exists at the most basic level of source verification.

For enterprise buyers using consulting reports to inform AI investment decisions: the credibility of a source does not substitute for the verifiability of a claim. When a report cites a named organization’s AI implementation, the implementation should be independently verifiable before it anchors a business case.

Source: The Decoder

Content created with AI assistance and reviewed for accuracy.