KPMG Withdraws AI Report After UBS, NHS and Others Debunk Fabricated Cases

KPMG Withdrew AI Report After UBS, NHS, and Others Disputed Fabricated Case Studies

KPMG withdrew the report "Rethinking Excellence in the Age of Agentic AI" — after four major organizations from different countries and sectors publicly disputed descriptions of their AI projects included in the report.

What Happened to the Report

The KPMG report was aimed at a corporate audience and focused on best practices for implementing agentic AI — systems capable of autonomously completing multi-step tasks. It included case studies from major companies and government structures, ostensibly as examples of successful implementation of new technologies. Such research from Big Four consultants is traditionally used as an authoritative reference point for corporate technology adoption decisions.

The problem came to light after the Financial Times contacted the organizations mentioned in the report. Representatives of these companies told the publication that they learned about their participation in the report only after it was published — and were extremely surprised. The described case studies did not match either the actual projects of the organizations or their official positions on artificial intelligence.

After a wave of denials, KPMG withdrew the report without detailed commentary. The company did not publicly explain how unreliable information about specific organizations ended up in an official document.

Who Disputed Their Case Studies

Four organizations from completely different sectors filed complaints against KPMG:

• UBS — one of the world's largest banks stated that the described AI initiatives do not pertain to them
• NHS (National Health Service of the United Kingdom) — reported that its case study was presented inaccurately or distorted
• Swiss Federal Railways — did not recognize the AI projects attributed to them
• Transport for London — also disputed the description of its AI activities in the report

Notably, all four organizations issued denials independently of each other — on their own initiative, without coordination. This rules out the theory of a single misunderstanding. It is also characteristic that none of the organizations questioned individual details — each stated that the described case studies simply did not pertain to them. This is a fundamental difference between inaccuracy and outright fabrication.

Hallucinations in an AI Report

It is hard not to notice the irony of what happened. The report, dedicated to "excellence" in the age of agentic AI, itself suffered from a phenomenon that remains the main problem of modern language models — hallucinations. This term refers to the confident reproduction of non-existent facts, names, quotes, and descriptions. If AI tools were used to gather or summarize case studies without proper verification, this explains the result: convincingly written but completely fabricated success stories.

It is precisely because hallucinations sound plausible and contain specific details that they are extremely difficult to detect without careful factual verification of every claim.

"The description of our work with AI does not correspond to reality," the

Financial Times quotes representatives of several of the mentioned organizations.

The scandal damages KPMG's reputation as a source of reliable analytics. Reports from the Big Four are traditionally considered verified — which is why they are cited in company strategies and presentations to boards of directors. If such reports can contain fabricated case studies, this changes how much they should be trusted without independent verification.

What This Means

The KPMG incident is a symptom of a broader problem. The hype surrounding agentic AI creates enormous pressure on analytical teams: publish materials faster than they can verify data. For any organization, this is a reminder: it is worth carefully monitoring how its name and projects are presented in industry reports and AI marketing. Distorted facts, once they make their way into a respected report, quickly begin to take on a life of their own — they are cited and referenced. One needs to respond before that happens.