GPTZero Investigation Reveals KPMG Report Riddled with AI Hallucinations

Summary

A new investigation by GPTZero, an AI detection software company, has exposed extensive AI hallucinations in KPMG's October 2025 report on agentic AI. The Big Four consulting firm's flagship report 'Total Experience: Redefining Excellence in the Age of Agentic AI' contained 40 out of 45 fabricated citations, with GPTZero coining the term 'vibe citing' to describe how generative AI tools accidentally create fake references or heavily paraphrase titles. The investigation also found that about half of the factual claims supported by these citations appear to be false or misattributed, with the LLM making persistent errors such as crediting blog posts to their subjects rather than authors, citing a 2019 railway press release as evidence of 2024-era agentic AI, and contradicting KPMG's own CEO Outlook published the same month.

The findings represent a broader crisis in professional services firms' use of AI without adequate human oversight. This is not an isolated incident—Deloitte issued refunds for an Australian report with AI errors, Pinsent Masons faced judicial criticism over AI-generated letters with false legal information, Sullivan & Cromwell apologized for hallucinated content in high-profile case filings, and EY had to withdraw a study with fake footnotes. The flawed statistics and claims from KPMG's report have already been recycled by industry publications and incorporated into major LLMs like ChatGPT and Gemini, demonstrating how hallucinations propagate through information ecosystems at scale.

• Professional services firms across the Big Four (KPMG, Deloitte, EY) and elite law firms have now faced public incidents involving AI-generated errors, signaling systemic failures in AI governance

Editorial Opinion

This investigation reveals a critical failure in how professional services firms are adopting generative AI—the assumption that machine-generated content only requires cursory review rather than rigorous human validation. The fact that KPMG's hallucinated research has already been ingested by mainstream LLMs like ChatGPT and Gemini underscores how AI-generated errors can corrupt entire information ecosystems, turning a single firm's negligence into infrastructure-level misinformation. Until professional services implement mandatory human verification of AI-generated citations, claims, and sources, these incidents will continue to erode both organizational credibility and trust in AI systems themselves.