A Report On The Benefits Of AI Was Reportedly Full Of AI Hallucinations
KPMG pulled a report on agentic AI after investigators found many false citations and misstatements. The paper itself appears to have been partly generated by AI.
Intelligence analysis by GPT-5.4 Mini
KPMG’s report on the value of agentic AI is drawing scrutiny after GPTZero and the Financial Times found fake or misleading citations and examples. The episode highlights how AI-generated research can undermine trusted corporate publishing when fact-checking is weak.
KPMG wrote a big report about AI, but many of the examples and references were wrong. It is like a school paper that cites books that do not exist and says a toy can do things it cannot.
Analysis
What happened
KPMG published a report in October titled Total Experience: Redefining Excellence in the Age of Agentic AI. The paper discussed how companies use AI to improve customer experiences, but investigators from GPTZero later found that it was filled with hallucinations, inaccurate claims, and fake footnotes.
The citation problem
According to GPTZero, only five of the paper’s 45 citations accurately pointed to real sources. The investigators said 28 citations paraphrased titles or added fake pieces to legitimate sources, while 12 were too vague to verify. GPTZero described this kind of AI-made fake referencing as “vibe citing.”
Bad examples in the paper
The report also appears to have misrepresented several companies’ AI use. One example claimed Emirates had launched a mobile chatbot called Sara that could talk to passengers and change flights, but the article says Sara was actually a mobile assistant launched in 2023 and did not have those booking powers. KPMG also said UBS had integrated agentic AI into investment advisory, risk management, and compliance monitoring; UBS told the Financial Times the claim was factually incorrect. Another example said Swiss Federal Railways had AI agents for trip planning and booking, which an SBB spokesperson said was not accurate.
Why this is a bigger issue
GPTZero estimated that about half the claims in the report were fake or misattributed, possibly because an AI research tool overdelivered on a request for examples of agentic AI in the wild. That is the real danger here: a report from a major firm can become a source that others trust, quote, and reuse. If the original is wrong, the errors can spread into later research and even into future AI-generated material.
KPMG told the Times that it takes accuracy seriously and said it was reviewing what happened. The firm has since pulled the paper.
Key points
- KPMG pulled a report on agentic AI after questions about fake and inaccurate citations.
- GPTZero said only 5 of 45 citations in the paper were accurate.
- The report mischaracterized examples involving Emirates, UBS, and Swiss Federal Railways.
- GPTZero warned that fake AI citations can spread bad information into later research.
- KPMG said it takes accuracy seriously and is reviewing what happened.
If KPMG tightens its review process, the incident could lead to better checks on AI-assisted writing and more reliable corporate research. It may also push companies to be clearer about when AI tools are used and how their output is verified.
If firms keep publishing AI-assisted reports without strong fact-checking, false claims could keep spreading through trusted business material. That would make it harder for readers, researchers, and AI systems to know what is real.
