The Big Four consultancy selling AI adoption just got caught using AI that made up its own success stories.
The Summary
- KPMG published a report promoting AI benefits that included fabricated case studies, with bogus examples involving UBS and transit systems that never happened
- UBS explicitly denied the AI integration claims attributed to them, forcing KPMG to acknowledge the hallucinations
- The incident exposes the governance gap between AI deployment velocity and verification rigor at firms selling AI transformation
The Signal
KPMG's report on AI benefits contained hallucinated case studies that fabricated specific examples of AI adoption. The report claimed UBS had implemented certain AI integrations and cited transit system deployments that never occurred. These weren't minor errors or misattributions. They were complete fabrications generated by the AI tools KPMG used to produce the report itself.
UBS issued a denial of the claims made about their AI implementation, leaving KPMG in the uncomfortable position of having to admit their AI-generated content wasn't fact-checked. A consultancy that advises Fortune 500 companies on AI strategy just published AI-generated fiction about AI success stories.
"AI's efficiency gains are undeniable, but the risk of unverified outputs necessitates robust governance to prevent costly errors and maintain trust."
The irony cuts deep. KPMG sells AI transformation. They counsel boards on AI governance frameworks. They audit AI implementations. And they just demonstrated exactly what happens when you skip the verification step: your AI tells you what you want to hear, and you publish it as fact.
Here's what this incident reveals about the current state of enterprise AI:
- Speed-to-publish beats accuracy-of-claims at firms with reputational capital to burn
- Even sophisticated users confuse AI fluency with AI accuracy
- The consulting model (bill for insights, not fact-checking) collides with AI's tendency to confidently bullshit
The impact on investor trust and market dynamics extends beyond KPMG's embarrassment. If reports from major consultancies contain unverified AI outputs, what else in the information supply chain is synthetic? Which analyst reports are citing hallucinated data? Which due diligence memos are built on made-up case studies?
This isn't a bug in the AI. It's a bug in the process. The tools work exactly as designed: they generate plausible-sounding text based on patterns. The risk comes from treating that output as verified truth without the human-in-the-loop step that checks claims against reality.
The Implication
Every firm rushing to "leverage AI" for content production just got a playbook for what not to do. The solution isn't to stop using AI for research and writing. It's to build verification layers that match the velocity of AI output. That means source-linking every factual claim, maintaining audit trails for AI-generated sections, and assuming every AI output is directionally useful but factually suspect until proven otherwise.
For anyone building in the agent economy: this is your moat opportunity. The firms that solve AI verification, that build trust layers for synthetic content, that create attestation systems for AI-generated work—those are the infrastructure plays that matter. KPMG just showed why the market needs them.