KPMG has quietly pulled one of its own reports after multiple organizations said its claims about their AI usage simply weren’t true. According to TechCrunch AI, the professional services firm removed a report titled “Redefining excellence in the age of agentic AI” from its websites following an investigation into what appear to be AI-generated fabrications. The irony writes itself: a firm appears to have used AI to help produce a report about AI, and the AI made things up.
What happened
The report was published in October 2025. Research group GPTZero flagged a series of inaccuracies in it, and told the Financial Times those errors stemmed from AI hallucinations, as detailed in TechCrunch AI’s reporting.
Several named organizations pushed back on how the report described their work:
- UBS
- The UK’s National Health Service
- Swiss Federal Railways
- Transport for London
All four told the FT that the report’s claims about their AI usage were either untrue or misleading. That’s not a typo or a stray data point. That’s the central content of the report being challenged by the very institutions it cited.
What KPMG said
A KPMG spokesperson confirmed the firm took the report down while it runs its own investigation. “We expect all our people to follow our guidelines on the responsible use of AI, including human oversight to validate content and verify independent sources,” the spokesperson said.
Read that line again. The firm is pointing to its own policy on human oversight, which is exactly the step that appears to have failed here. Guidelines don’t catch hallucinations. People checking the work do.
Why this matters
What stands out is that this isn’t an isolated slip. Last month, EY withdrew a report on loyalty rewards programs that appeared to include fake footnotes and AI hallucinations. Two of the world’s largest professional services firms, both pulling published research over AI fabrications, inside a few weeks of each other.
This is significant because these firms sell trust. Their entire business rests on clients believing the analysis is rigorous, sourced, and verified. When the output contains invented claims about named institutions, the damage isn’t just one bad PDF. It’s the credibility of the research function itself.
The pattern points to a specific failure mode that practitioners should recognize:
- AI drafts plausible-sounding content fast.
- The output reads polished, so reviewers trust it.
- Nobody independently verifies the specific factual claims.
- The errors only surface when an outside party, or the subject of the claim, calls it out.
Hallucinations don’t look like errors. They look like confident, well-formatted sentences. That’s what makes them dangerous in professional work, and that’s why a spell-check mindset doesn’t catch them.
What to take away
If you’re using AI to produce anything that carries your name or your firm’s name, treat these withdrawals as a warning shot. A few practical guardrails:
- Verify every factual claim, statistic, and quote against a primary source before publishing. Especially claims about named third parties.
- Check footnotes and citations individually. Fabricated references are a known hallucination signature, and they’re the EY case in a nutshell.
- Build human review into the workflow as a required gate, not an optional courtesy.
- Assume polish is not proof. Fluent text and accurate text are different things.
The firms aren’t abandoning AI over this, and they shouldn’t. The tools are useful for drafting, structuring, and speeding up research. What both cases show is that the verification layer is not negotiable. Cut it, and you ship fabrications under a trusted brand.
Expect more of these retractions before the industry tightens up. As AI-assisted writing spreads through consulting, finance, and corporate research, the firms that win won’t be the ones moving fastest. They’ll be the ones who verify what the machine wrote before it goes out the door. You can find the full details at the original source.