KPMG’s big AI report promised a roadmap to excellence. Instead, it delivered fiction. The October 2025 study on agentic AI use was riddled with fabricated data and bogus citations.

AI detection firm GPTZero exposed that 40 out of 45 citations in the report were fake. Many references blended real sources with invented ones or mangled titles beyond recognition—a phenomenon dubbed “vibe citing.”

The report claimed Swiss bank UBS deploys AI agents for investment advice and risk management. UBS called that “factually incorrect.” Swiss Federal Railways and Transport for London denied using AI agents as described. NHS Greater Manchester’s AI claims were twisted from unrelated press releases.

KPMG pulled the report after these errors surfaced. Yet, the damage was done. Industry publications and even AI chatbots like ChatGPT started recycling the false claims, spreading misinformation across the web.

Hallucinations from Trusted Advisors

Big consultancies have become repeat offenders. Deloitte, EY, and prestigious law firms have all issued AI-tainted reports recently. Deloitte refunded part of a government contract after AI-generated errors. EY retracted a study with fake footnotes. Sullivan & Cromwell apologized for AI hallucinations in court filings.

These firms sell themselves as AI experts guiding responsible adoption. Yet, their own reports suffer from the very flaws they warn clients to avoid. The common thread: a lack of human oversight before publication.

KPMG’s report even contradicted its own CEO survey data. It cited 55% of executives prioritizing AI investment, while its official 2025 CEO Outlook stated 71%. Such glaring inconsistencies reveal careless fact-checking.

GPTZero’s CEO Edward Tian warned this cycle “poisons the well of information.” When trusted firms publish falsehoods, it increases the risk of “second-hand hallucinations” multiplying across industries.

This episode underscores a harsh truth: AI hallucinations are not just a technical glitch. They reflect the failure of governance, editorial rigor, and accountability in the rush to tout AI’s promise.

KPMG admitted its internal processes failed. The author of the report confessed no one double-checked citations or claims before publishing. The firm now promises stricter guidelines on AI use and human review.

Until such safeguards are standard, AI-generated content—even from the biggest players—demands skepticism. The allure of AI’s efficiency risks drowning truth in a sea of confidently stated nonsense.

For companies looking to adopt AI responsibly, the takeaway is clear. Trust but verify. Otherwise, you risk outsourcing critical thinking to an algorithm that hallucinates facts and invents reality.