KPMG PULLS REPORT AFTER AI HALUCINATES

KPMG published a report called in October 2025 and then had to pull it after a number of well-known organizations said the report's claims about their AI usage were flat-out wrong. UBS, the UK's National Health Service, Swiss Federal Railways, and Transport for London all told the Financial Times that the report either described them using AI in ways they never did, or presented misleading accounts of their actual work. The research group GPTZero identified a series of inaccuracies in the document, and told the FT that the errors stemmed from AI hallucinations.

KPMG appears to have used AI to help write a report about AI, and then got caught when the hallucinated case studies were fact-checked by the organizations they supposedly described.

WHAT THE REPORT CLAIMED & WHO PUSHED BACK

The report was positioned as a forward-looking analysis of how — autonomous AI agents that can act without direct human intervention — is reshaping corporate excellence. It included several high-profile organizations as examples of real-world deployment. The problem is that those organizations say the examples were fabricated.

UBS, one of the world's largest wealth managers, rejected KPMG's characterization of its AI work. The NHS, hardly a shy adopter of new technology but also a highly regulated institution, said the claims about its AI usage were untrue. Swiss Federal Railways and Transport for London, two of Europe's largest public transport operators, both denied the report's descriptions of their AI implementations.

The FT broke the story, and GPTZero — a company that builds tools to detect AI-generated content — subsequently analyzed the report and flagged multiple instances where the document appeared to have hallucinated citations and use cases. GPTZero told the FT that the pattern was consistent with generative AI models inventing plausible-sounding but false details, a phenomenon well known to anyone who has spent more than ten minutes with ChatGPT.

THE IRONY IS THICK ENOUGH TO BE ITS OWN CASE STUDY

KPMG is one of the Big Four accounting and advisory firms. These firms bill corporations millions of dollars a year for advice on digital transformation, AI strategy, and risk management. If KPMG could not produce a credible report about AI without its own AI hallucinating fake examples, what does that say about the advice it is selling to clients?

A KPMG spokesperson said the firm removed the report from its websites while conducting its own investigation. That is the standard playbook when a report is publicly embarrassed: pull it, investigate, wait for the heat to die down, then issue a softer statement about lessons learned. What the investigation will need to answer, and what matters to the industry, is how the hallucinated material got through the editorial process. Was there an editor? Was the report run through a human fact-check against the organizations named? If not, that is not a technical failure, it is a process failure. And a process failure at a Big Four firm is a credibility problem for the entire consulting industry.

THIS IS NOT THE FIRST TIME

EY, another Big Four firm, withdrew a report on loyalty rewards programs earlier in 2025 that appeared to include fake footnotes and AI hallucinations. The EY incident followed the same pattern: a report that looked polished, cited real companies and publications, but the footnotes led nowhere or cited articles that never existed.

Two Big Four firms, two AI-hallucinated reports, both pulled after media scrutiny. That is not an anomaly anymore. That is a pattern. Consultants are under enormous pressure to produce thought leadership content — reports, white papers, blog posts, LinkedIn thought pieces — to maintain their market positioning. Generative AI makes that content cheap and fast. The temptation to let the AI write the first draft and skip the human review is obvious. The consequence is that when the AI gets it wrong, the firm looks incompetent in its own advertised area of expertise.

KPMG AND EY BOTH SELL AI ADVISORY SERVICES

Both are presumably telling clients to deploy generative AI responsibly: to have human oversight, to fact-check outputs, to maintain governance frameworks. Yet both have now been publicly caught failing to follow that advice in their own public-facing work. The gap between what they preach and what they practice is now a matter of public record.

The damage is not just reputational. Clients who pay for AI consulting are going to start asking harder questions about how the consultants themselves are using these tools. If KPMG could not be trusted to write a simple survey report without hallucinating, can they be trusted with a client's AI implementation strategy? The answer, for now, is that they have made it much harder for themselves to say yes.

THE LESSON: WE ARE NOT THERE YET FOLKS.

The KPMG episode is a useful reminder that generative AI is not a research tool. It is a text generator that produces plausible-sounding strings of words. When it is asked to produce a case study about UBS using agentic AI, it does not look up what UBS actually did. It builds a sentence that looks like what a case study should look like. That sentence may be correct by coincidence, but it is never correct by design. And when the organization in question reads the sentence, they will instantly know it is wrong.

Consultancies that use AI to write reports without rigorous human fact-checking are putting their entire credibility at risk for two days of saved labor. That calculation does not hold up to even the simplest cost-benefit analysis. But the fact that it happened twice at two different Big Four firms in the same year suggests the industry has not yet learned this lesson.

KPMG's investigation should be interesting. Whether they release its findings publicly will tell us how seriously they take the problem. If the report resurfaces in a corrected form with a mea culpa, that would be a mature response. If it simply disappears and everyone pretends it never happened, that would be a different kind of statement.

For everyone else watching: if you are producing content about AI, do not let the AI write the examples. And if you are paying for AI consulting, ask the firm how they made their own report before you let them touch your strategy.