KPMG publishes AI agents report riddled with hallucinations: invented quotes and nonexistent case studies

KPMG, one of the largest consulting firms in the world, released a report on artificial intelligence agents in October 2025. The problem is that the document itself appears to have been written with AI: it contains fabricated quotes and references to case studies that do not exist.

What was found in the document

Researchers who examined the KPMG publication recorded several types of errors typical of AI hallucinations. The text included quotes attributed to real experts who, however, never said them. Some references led to studies that do not exist, that is, to materials not listed in any archive. Here is the typical set of problems the reviewers identified:

• Quotes from real people with invented content
• References to nonexistent case studies
• Statistics without verifiable sources
• Cases of real companies with fictitious details
• Mentions of studies that cannot be found in public databases

Such errors are well known to AI specialists: these are classic language model hallucinations, the confident generation of plausible but invented content.

Why this hurts for a "Big Four" firm

KPMG is part of the "Big Four" global audit and consulting firms, alongside Deloitte, PwC and Ernst & Young. Their key competitive advantage is accuracy, reliability and authority. Corporate clients pay tens of thousands of dollars for research precisely because they trust its quality.

The context adds particular irony: KPMG actively sells clients consulting services on AI adoption and warns about the risks of carelessly used AI tools. The scandal showed that the company itself fell into the same trap it warns others about.

How does this even happen?

The exact circumstances have not been disclosed, but the scenario is quite typical. An analyst uses a language model to speed up writing a study, which is normal practice in 2025. The problem arises at the verification stage: if the author does not manually verify every quote and every reference, hallucinations go unnoticed. With demand for AI expertise high, deadlines are often tight, and the temptation to delegate routine work to the model is strong. But for material where reputation depends on accuracy, skimping on verification comes at a high price.

"When a company that advises on AI risks releases a report with AI

hallucinations, it is the best illustration of the risk itself."

What this means

The KPMG incident is a warning for the whole industry: using AI to produce professional reports requires mandatory manual verification of quotes and sources. A brand's reputation is no substitute for fact-checking, and the greater the trust in the author, the more damaging the consequences when that trust is not justified.