KPMG Published an AI Report Stuffed With Fake Citations. The Firm Has Since Pulled It.

What KPMG Published

In October 2025, KPMG released a report titled Total Experience: Redefining Excellence in the Age of Agentic AI. The paper was meant to show how companies are deploying AI agents to serve customers. KPMG is not a fringe think tank — it is one of the four largest professional services firms on the planet, alongside Deloitte, PricewaterhouseCoopers, and Ernst & Young. These firms produce research that gets cited by other researchers, journalists, and policymakers who assume the sourcing is solid.

It was not.

The Numbers Are Ugly

GPTZero, the company behind an AI content detection tool, investigated the report and published its findings. The Financial Times independently verified the results.

Of 45 citations in the paper, only five accurately pointed to real sources, according to GPTZero. Twenty-eight citations paraphrased titles or inserted fabricated components into real sources. The remaining 12 were worded so vaguely that investigators could not determine whether the underlying sources existed at all.

Approximately half the specific claims in the paper were either fake or misattributed. GPTZero attributed this to an AI research tool that over-complied with a prompt asking it to find examples of "agentic AI in the wild" — generating plausible-sounding examples that did not exist.

Real Companies, Fake Claims

The fabrications were not abstract. KPMG named real, well-known companies and invented capabilities for them.

The report claimed Emirates launched a mobile chatbot called Sara that could hold conversations with passengers and modify their flight bookings. According to Engadget's reporting on the GPTZero findings, Sara was a mobile assistant launched in 2023, not an AI-powered chatbot, and it had no ability to change passenger bookings.

KPMG stated that UBS had integrated agentic AI across its "investment advisory, risk management and compliance monitoring." UBS told the Financial Times directly that this was "factually incorrect."

Swiss Federal Railways (SBB) was listed as running AI agents that help passengers plan, book, and optimize trips based on real-time conditions and carbon impact. An SBB spokesperson called that "not accurate."

Three named companies, three corrections. None of them were small or obscure.

"Vibe Citing"

GPTZero CEO Edward Tian coined a term for what happened: "vibe citing." It describes AI models generating references that feel credible — plausible-sounding titles, author names, publication years — without any real document behind them.

Tian warned that when firms with KPMG's institutional credibility publish error-riddled research, it does not stay contained. Other papers cite KPMG. Journalists quote KPMG. Executives make decisions based on KPMG. Fake claims from a trusted source become second-hand AI hallucinations embedded in the broader information ecosystem. That is a different problem than a random chatbot making things up.

The Strongest Counterpoint

KPMG is not unique here, and framing this purely as an AI problem misses half the story. AI tools do not publish reports. Human editors, partners, and communications teams at KPMG reviewed and approved this paper before it went out. Critics of AI panic sometimes argue that the real culprit is institutional corner-cutting — organizations deploying AI drafting tools without building adequate human verification layers on top.

The AI did not publish the report. KPMG did. The accountability sits with the people who signed off on it. Whether the failure was human laziness, misplaced AI trust, or both, the result was the same: a widely circulated paper with fake evidence, under a trusted brand name, promoting the very technology it failed to use responsibly.

KPMG's Response

A KPMG spokesperson told the Financial Times that the company "takes the accuracy and integrity of its published content seriously." The firm has pulled the report and is "reviewing the circumstances surrounding its publication."

That review has not produced any public findings as of June 13, 2026. No one at KPMG has been named publicly as responsible for the paper's production or approval.

The Unresolved Question

KPMG's internal review is the concrete next step. The unanswered question is what it will produce. Will the firm disclose specifically where in its editorial process the AI-generated content was treated as verified research? Will it identify whether this was an isolated incident or a pattern across other AI-themed reports? GPTZero's Tian has made the stakes clear: if firms of this caliber do not answer that question with specifics, the credibility damage extends well beyond one pulled PDF.