KPMG withdrew a report on artificial intelligence after some of the largest organizations in the world stated that what the document said about them was made up.
The report, titled "Redefining Excellence in the Age of Agentic AI" and published in October 2025, included case studies on how large companies and institutions were adopting AI in their day-to-day operations. The problem: the people who know those companies best — the companies themselves — said it wasn't true.
Who Denied KPMG's Claims
The Financial Times contacted the organizations named in the report. The responses were blunt:
- UBS stated that the claims about its AI use were false
- NHS (the UK's national health service) denied what was reported
- Swiss Federal Railways said the claims were misleading
- Transport for London did the same
KPMG removed the report from its website while it conducts an internal investigation.
The Paradox: An AI Report Written with AI
The most embarrassing part of the story comes from GPTZero's analysis, which identified the likely cause of the inaccuracies: AI hallucinations. It appears KPMG used artificial intelligence to help write a report about artificial intelligence — and that the model invented details about the case studies without anyone verifying them.
The result is a document describing how large organizations used AI in ways those organizations never actually did.
It's a story that's both funny and worrying at the same time.
Why This Story Matters
KPMG isn't some minor blog. It's one of the four largest consulting and auditing firms in the world, with clients in every sector. When a firm of this caliber publishes a report full of AI-invented claims — about AI — the message that reaches the market is clear: no one is immune from AI errors if generated content isn't verified.
This case raises questions every company using AI to produce content should ask itself:
- Who verifies that what the AI writes is true?
- Is there a human fact-checking process before publication?
- Are cited sources or claims made about third parties confirmed?
The Practical Lesson
AI is a powerful tool for speeding up writing, research, and synthesis. But a language model can't tell the difference between what's true and what sounds plausible. It generates fluent, structured, convincing text — regardless of factual accuracy.
For professionals using AI in their workflows, the rule is simple: anything the AI states as a specific fact about a third party needs to be verified before publication. Not as a theoretical best practice — as a minimum standard of professional responsibility.
KPMG learned this in the most public way possible.
KPMG's report withdrawal isn't a story about AI going wrong. It's a story about negligent use of AI by people who should have known better. AI works well when supervised by someone who knows what they're looking for — and fails spectacularly when left alone to invent case studies about UBS and the NHS.