KPMG Published a “Future of AI” Report Where 40 of 45 Citations Were AI Hallucinations — Then Quietly Pulled It.
One of the largest consulting firms in the world published a report on the future of AI and customer experience in which 40 of 45 citations appeared to be at least partially AI-hallucinated, according to AI-detection firm GPTZero. The report claimed Japan's JR East was using agentic AI for customer service — citing a 2019 press release, years before agentic AI existed. It claimed Austrian utility Verbund ran AI agents in households for real-time analytics — a claim with no supporting evidence.
Many citations linked to real studies but with wrong titles, authors, or dates; others couldn't be matched to anything at all. GPTZero's conclusion: an LLM research tool “conflated sources, exaggerated claims, and injected references to agentic AI that were copied into the final report without verification,” and “no human at KPMG double-checked the citations.”
The kicker: before removal, the report was cited by multiple trade publications and a major Czech newspaper — and is now being surfaced as a “resource” by ChatGPT and Google Gemini, laundering fabrications into the very systems it described. This is the contamination loop. A trusted firm let an AI tool fabricate its sources, published it under its name, and those fabrications now feed back into the AI systems people trust for answers.
Human-in-the-loop wasn't bypassed by malice — it was simply skipped. The brand was the only verification anyone relied on, and the brand didn't read its own report. KPMG's statement: “We expect all our people to follow our guidelines on the responsible use of AI, including human oversight to validate content.” HITL Score: 9/100 — human oversight at deployment 2/25 (an LLM tool's output was copied into a published report unverified), ongoing monitoring 1/25 (caught by an outside party, GPTZero, not KPMG), incident response 4/25 (report pulled and “reviewing circumstances,” after wide circulation), accountability 2/25 (the firm points to its own guidelines that were plainly not followed).