PwC Published AI Reports Riddled With AI Hallucinations

PwC, the consulting firm that has spent months urging clients to adopt artificial intelligence, has itself published a series of AI reports written largely by artificial intelligence and full of fabricated facts. An investigation by Financial Times journalists, based on analysis from the firm GPTZero, found fictional citations, nonexistent studies, and products that no one outside the company itself has ever seen in four PwC Middle East documents.

What the Investigation Found

The most telling example is the report 'Transforming Governance,' which promotes PwC's proprietary tool called Citizen Pulse. The document claimed the solution was used by the governments of Denmark, Saudi Arabia, the United States, and Australia. GPTZero found no public trace that such a product had ever been deployed, or that the government agreements mentioned ever existed.

In another report, on electromobility and autonomous vehicle cybersecurity, researchers found a footnote pointing to an alleged scientific study on air quality in Riyadh. It could not be found either in the journal where it supposedly appeared or in the body of work of the authors credited, pointing to outright fabrication.

A Teenager's Medium Post as a Source

One bizarre example concerns JPMorgan's automation initiative. PwC cited a post by a teenage blogger on Medium with just 280 followers, describing an event from 2017, five years before ChatGPT's public debut. GPTZero researcher Paul Esau also flagged a claim that recurred across the reports, that human error accounts for 90 percent of road accidents. The same claim appeared in the documents once with one source, once with no citation at all, and once with two different sources, a telltale sign of automated text generation.

PwC Middle East takes the accuracy of its published research seriously and is updating a limited number of source citations - a PwC spokesperson, quoted by the Financial Times

The Big Four's Recurring Problem

PwC is not the first consulting firm caught in this kind of slip-up. Similar accusations previously hit KPMG, where one AI report had only 5 of 45 footnotes actually backing up the claims they were attached to, as well as EY Canada, which in May 2026 withdrew a document containing false data, and Deloitte, whose report for the Australian government, containing nonexistent citations, ended with a partial refund of its fee. The pattern is the same every time: firms that sell clients on AI deployments are themselves publishing promotional materials written largely by language models, without proper fact-checking.

Why It Matters for Companies in Poland

The affair strikes at the credibility of so-called thought leadership, the expert reports on which major consulting firms build their reputations and sell further consulting contracts. If materials promoting AI deployments turn out to be themselves the product of unchecked AI generation, corporate clients, including companies in Poland that use Big Four consulting services, gain a concrete argument to demand transparent information about how such reports are produced and proof that the cited deployments actually exist.

The case also lands at a moment when Polish companies increasingly base business decisions on industry reports and AI benchmarks, often without any way to independently verify the foreign sources cited in them. The PwC example shows that even materials bearing the brand of a global consulting leader now require the same skepticism applied to anonymous online content.

The Financial Times notes that the irony is compounded by the fact that PwC's US leadership had previously warned clients that artificial intelligence merely exposes pre-existing flaws in organizational processes. The firm's own reports have become, in this context, a textbook example of that very thesis in practice.

PwC has not disclosed how many additional reports will be reviewed beyond the four identified by GPTZero, nor whether the firm will introduce systematic procedures for labeling AI-assisted content before publication. Until such mechanisms exist, experts quoted in the investigation say similar slip-ups across the consulting industry remain a matter of time.