Glonce

PwC Middle East allegedly published AI-generated reports with fake sources

miscThe Glonce editorsThe Decoder🔥73Glonce rating73Impact70Novelty60Relevance90Surprise65Players65Credibility80
PwC Middle East allegedly published AI-generated reports with fake sources

AI-detection company GPTZero examined four reports published by PwC Middle East between 2024 and 2026 and says it found fabricated sources and false claims running through them. The worst case is a report called "Transforming Governance," which GPTZero rates as 84 percent likely to have been entirely AI-generated. That report is used to promote a PwC product called "Citizen Pulse," and it claims that governments in Denmark, Saudi Arabia, the United States and Australia rely on the product. GPTZero says no evidence backs those claims. Across the four reports, the ones with the highest AI-generation scores also carried the most false sources, a pattern GPTZero labels "Vibe Citing": references that are attached loosely, often missing correct titles, URLs or authors, and that do not actually support the statements they are cited for. GPTZero notes the same pattern shows up in academic papers. PwC Middle East told the Financial Times that it takes accuracy seriously and is "updating a limited number of supporting citations," but did not explain how the errors made it into the published reports. This follows GPTZero's earlier findings of fabricated sources in reports from KPMG, Deloitte and Ernst & Young, making PwC the latest of the Big Four to be flagged.

Key facts

  • GPTZero found fabricated sources and false claims in four PwC Middle East reports published between 2024 and 2026.
  • The report "Transforming Governance" is rated 84 percent likely to be entirely AI-generated, the highest score of the four.
  • That report claims governments in Denmark, Saudi Arabia, the U.S. and Australia use PwC's "Citizen Pulse" product; GPTZero says no evidence supports this.
  • GPTZero terms the pattern "Vibe Citing": loosely attached references, often missing correct titles, URLs or authors, that do not back the claims they accompany.
  • PwC Middle East told the Financial Times it is "updating a limited number of supporting citations" but did not explain how the errors occurred; GPTZero had previously flagged fabricated sources at KPMG, Deloitte and Ernst & Young.

Why it matters

This is the fourth of the Big Four accounting and consulting firms that GPTZero has caught publishing reports with fabricated sources, after KPMG, Deloitte and Ernst & Young. It suggests unchecked generative AI use has become a recurring failure mode in high-profile professional-services publishing, not an isolated incident, and it lands on a report that was also being used to market a specific PwC product with unverified client claims.

Who it affects

PwC's Middle East practice and its reputation for rigor; the governments named in the report (Denmark, Saudi Arabia, the United States and Australia), whose alleged use of "Citizen Pulse" GPTZero says is unsupported; and readers or clients who relied on the four reports' citations as accurate.

How to use it

For anyone citing consulting-firm or corporate reports, the story is a reminder to check the underlying sources directly rather than trust citations at face value, especially in reports built for marketing a specific product. GPTZero's "Vibe Citing" pattern (correct-looking but wrong or unsupported references) is a concrete thing to check for: verify that a cited title, URL and author actually exist and actually support the claim attached to them.

How solid is it

The claims rest on GPTZero's own AI-detection scoring and its manual check of citations, and the 84 percent figure is GPTZero's model estimate rather than a confirmed admission. PwC Middle East has responded to the Financial Times acknowledging it is correcting some citations, which lends some corroboration, but it has not confirmed AI authorship or explained the cause, and the source text itself notes that whether this is really an "AI problem" is a separate, open question.

Risks and caveats

AI-generation detection scores are probabilistic and not proof of authorship; an 84 percent likelihood is not a confirmed fact. PwC has not detailed how the errors occurred, so it isn't established whether AI drafting, human error, or some combination is responsible. The claim that Citizen Pulse is used by the four named governments is described as unsupported by GPTZero, not as independently disproven by another party.

“Whether this is really an AI problem, though, is a separate question.”

— The Decoder