KPMG Pulls AI Report After Finding False Case Studies and Fabricated Citations

KPMG has withdrawn a report called Redefining Excellence from its websites after discovering the document contained AI-generated errors, including false citations, invented case studies, and claims about AI use that organizations had never actually made.

The errors were identified by GPTZero, a research group that detects AI-generated text, and later confirmed by the Financial Times. Among the problems were case studies that falsely described AI projects at UBS and various health and transit systems. When UBS learned about the false claims, the bank asked KPMG to remove any reference to it. Multiple other organizations named in the report said the descriptions of how they used AI were simply incorrect.

KPMG removed the report while investigating how it was released with these errors, according to The Register. The company has not announced a timeline for completing the investigation.

What Went Wrong

Large language models — the AI systems behind tools like ChatGPT — have a known flaw called hallucination. The models generate text that sounds confident and detailed but is often factually wrong. They sometimes invent facts, misquote sources, or make up examples that never happened. This is one of the most talked-about risks in enterprise AI adoption, and it's something KPMG explicitly advises clients about.

The irony is sharp: a major professional services firm published a report about AI risks that became a live example of the very problem it was supposed to address.

The real issue here, though, is not that an AI system made mistakes. It is that KPMG apparently did not have adequate human review before the report went public. AI models produce wrong outputs all the time. The difference between responsible use and negligent use is what happens next: someone needs to verify the results before they reach the outside world. In this case, the verification process failed. No one caught the fabricated company names and false attributions before publication.

The Trust Problem

KPMG's entire business relies on trust. Clients pay for audit, advisory, and risk services because they expect rigorous accuracy and fact-checking. A published report that invents case studies and attributes them to real, named companies — companies that then had to contact KPMG to demand corrections — directly undermines that value proposition. The reputational damage is concrete and material.

The broader industry context matters too. Consulting and research firms have moved quickly to use generative AI in their report production, often under pressure to publish faster. These tools can genuinely help with drafting, summarizing data, and pulling information together from multiple sources. But the oversight mechanisms — editorial review, fact-checking, legal review of any claims about real companies — have not always caught up with how fast firms are deploying these tools. This incident will likely force tighter controls across the consulting sector, not just at KPMG.

What to Watch For

Anyone working with AI-assisted research and advisory reports should apply basic verification to any specific claims. This means: independently check any named companies, specific numbers, and case study details before relying on the document. LLM-generated writing is often so smooth and confident-sounding that false claims are easy to miss when you read quickly. That makes the verification step essential, not optional.

KPMG has not disclosed which specific AI tools produced the report, or whether the hallucinations came from the initial drafting stage or somewhere else in the production workflow. Those details matter. They would help other firms understand what kind of failure this was — and help them audit their own AI-assisted publication processes in response.