KPMG Pulled an AI Report — the AI Made It Up

I build with Claude Code every day, so I'm the last person who'll tell you to put the AI down. But this story made me wince — the useful kind of wince, the one that sends you back to double-check your own work. KPMG, one of the Big Four accounting firms, quietly pulled a report it had published called 'Redefining excellence in the age of agentic AI.' The reason: a chunk of the 'facts' inside it appear to have been hallucinated by AI.

The report went out in October 2025. A research group called GPTZero went through it and flagged multiple inaccuracies, pinning them on AI hallucinations. The document claimed specific organizations were using AI in specific ways — and several of those named organizations, including the bank UBS, the UK's National Health Service, Swiss Federal Railways and Transport for London, said the claims about them were, in their words, 'either untrue or misleading.' When real institutions have to publicly say 'that thing the report said we did? We didn't,' there's nothing left to do but pull it. KPMG did, while it investigates. The Financial Times reported the retraction.

This isn't a KPMG problem — it's an everyone problem

And it's not a one-off. Just a month earlier, EY — another Big Four firm — withdrew a loyalty-rewards report that also turned out to carry fabricated footnotes and AI hallucinations. Think about who these companies are: the most process-heavy, review-everything, get-paid-to-check-other-people's-work firms on the planet. If invented sources can slide past their reviewers, they can absolutely slide past yours and mine. KPMG's own spokesperson basically admitted the failure was human, not robotic: '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.' Translation: the AI didn't break the rules. The person who skipped the verification step did.

Here's why this matters even if you've never opened a spreadsheet in your life. AI hallucinations are confident. A made-up statistic doesn't arrive flagged in red; a fake citation looks exactly like a real one, right down to the formatting. The danger was never the obviously-wrong answer — you catch those instantly. It's the plausible-looking detail you don't think to question, because everything around it checks out. That's how it slipped past a room full of KPMG reviewers, and it's the same way it slips into the email, the school essay, or the work summary you generate this week.

I want to be clear: I'm not anti-AI, I lean on it harder than almost anyone I know. But the firms in this story didn't get burned by using AI. They got burned by trusting it without looking. That's a free lesson handed to the rest of us, and it costs nothing to apply — just the thirty seconds it takes to check before you send.