PwC Middle East Revises Reports After AI Hallucinations and Fake Citations Exposed

PricewaterhouseCoopers Middle East has been forced to review and correct a series of corporate research reports after independent analysis revealed widespread artificial intelligence “hallucinations,” fabricated citations, and invalid web links. The flawed publications—issued between 2024 and 2026—were flagged by AI detection platform GPTZero before being confirmed by media outlets including the Financial Times.

The incidents highlight growing risks for top-tier professional services networks, commonly known as the “Big Four”—comprising PwC, Deloitte, EY, and KPMG. As accounting and strategy firms rush to position themselves as authorities on enterprise digital transformation, the use of unverified large language model outputs in client-facing “thought leadership” has repeatedly compromised research standards.

Evidence of unvetted AI generation was found embedded directly within document footnotes. In one cybersecurity report, a citation URL contained the tracking string “utm_source=chatgpt.com,” revealing that the source material had been copied directly from OpenAI’s web interface. Another report on regional business trends asserted that nearly 70% of Middle Eastern chief executives expect generative AI to disrupt their operating environment, yet the corresponding footnote linked to an article containing no reference to such a survey.

A report on electric vehicles and regional infrastructure repeatedly misattributed basic statistics. The publication cited a figure claiming human error causes 90% of traffic collisions three times across two pages, using two entirely different academic citations for two instances while leaving the third without any source. The same document included references to a Riyadh air quality study that could not be located in academic databases, alongside multiple broken hyperlinks.

In another case, PwC cited a post from a Medium user with roughly 280 followers to document an automated commercial loan review project at JPMorgan. While the report presented the case study as a contemporary AI success story, the bank’s actual software implementation took place in 2017—years prior to the release of consumer-facing large language models like ChatGPT.

Automated research tools often exhibit what computer scientists term “hallucination,” generating convincing references, URLs, or statistical relationships that do not exist in reality. GPTZero policy analyst Paul Esau noted that erratic citation patterns are a primary indicator of synthetic text, stating that chaotic signposting of source material is characteristic of unedited AI output.

The findings carry reputational risks for PwC, which actively markets advisory services helping corporate clients implement responsible governance frameworks for emerging technologies. Similar incidents have previously impacted rival firms EY and KPMG, both of which withdrew research papers flagged for synthetic text content by GPTZero.

Responding to the disclosures, PwC Middle East stated that it takes research accuracy seriously and confirmed it is in the process of updating supporting citations across the affected publications.