KPMG pulls agentic AI report after organisations dispute claims

Images

×

Image Credits:CFOTO / Getty Images Getty Images

techcrunch.com

a screenshot from the FBI's Kinetic Cyber Range video, showing the FBI's model town cross walk, including a business center, and a light post, inside a large warehouse space. techcrunch.com

SUQIAN, CHINA - MAY 19, 2026 - A netizen is checking KPMG logo on his mobile phone and KPMG webpage on his computer in Suqian City, Jiangsu Province, China on May 19, 2026. (Photo credit should read CFOTO/Future Publishing via Getty Images) techcrunch.com

Andy Jassy the CEO of Amazon speaks at the ceremonial ribbon cutting prior to tomorrow's opening night for the NHL's newest hockey franchise the Seattle Kraken at the Climate Pledge Arena on October 22, 2021 in Seattle, Washington techcrunch.com

‹ ›

KPMG has pulled an AI-themed report after multiple organisations named in it disputed claims about their use of artificial intelligence, according to TechCrunch and the Financial Times. The report, titled Redefining excellence in the age of agentic AI and published in October 2025, was taken down while the firm conducts an internal investigation. GPTZero, a research group that analyses AI-generated text, told the Financial Times it found several inaccuracies and attributed them to “hallucinations” — fabricated or incorrect statements produced by generative systems.

The episode lands awkwardly because the document was itself a marketing-style guide to deploying “agentic AI”, a category of tools sold on the promise that they can plan, execute and coordinate work with limited supervision. If a report meant to showcase institutional readiness cannot reliably describe whether large, brand-sensitive organisations are using the technology, it exposes a basic problem for the entire genre: the most valuable claims are often the easiest to publish and the hardest to verify. Naming household institutions signals credibility to prospective clients, but those institutions also have strong reasons to correct the record when a third party implies operational choices they did not make.

The Financial Times reported that UBS, the UK’s National Health Service, Swiss Federal Railways and Transport for London said the report’s descriptions of their AI usage were untrue or misleading. KPMG, in a statement cited by TechCrunch, said it expects employees to follow internal guidelines on responsible AI use, including human oversight, validation of content and verification of independent sources. That language is now part of the story: a global professional-services firm is explaining after the fact that its own process requires the very checks that appear not to have caught the disputed passages.

KPMG is not alone. TechCrunch notes that EY withdrew a separate report in May 2026 after it appeared to include fake footnotes and other AI-style errors. The pattern is less about any one tool than about how quickly “thought leadership” can be produced and distributed when the cost of drafting collapses but the cost of fact-checking does not. A glossy PDF can travel through client networks, conference panels and procurement meetings long before the organisations it references notice they have been drafted into someone else’s narrative.

For readers, the practical consequence is not that AI text sometimes gets things wrong — that is already widely understood — but that institutional publishing pipelines are starting to treat error correction as a reputational afterthought rather than a prerequisite. In industries where reports are used to justify spending, staffing and technology procurement, the difference between “drafted fast” and “checked thoroughly” is often the difference between a budget request that dies and one that becomes a multi-year contract.

KPMG has removed the report from its websites while it investigates what went into it. The disputed claims remain easiest to find in the screenshots and citations left behind.