Facepalm: PricewaterhouseCoopers, one of the world's four largest accounting firms alongside EY, KPMG, and Deloitte, has landed in hot water after it was found to have published at least four reports riddled with AI hallucinations, fabricated citations, and fake footnotes. The reports were issued by PwC Middle East between 2024 and 2026 and were flagged as AI-generated by GPTZero, an AI detector and plagiarism checker.
The AI hallucinations were confirmed by multiple leading publications, including the Financial Times. The so-called "Thought Leadership" reports were intended to articulate PwC's insights into some of the fastest-growing businesses in the Middle East, including agentic AI, public services, and electric vehicles. However, the bizarre turn of events suggests that the reports themselves were AI-generated.
One of the contentious reports claimed that an internal PwC survey found that nearly 70% of CEOs in the region believed generative AI would redefine their business landscape in the near future. However, a footnote in the document linked to an article that did not cite any such survey.
Another footnote, this time in a cybersecurity report, included a link containing the tracking tag "utm_source=chatgpt.com," leaving very little doubt about its source.
In one particularly egregious case, a report cited a Medium blog post by a teenager with 280 followers as the sole source for a "real-world success story" about an AI deployment initiative by JPMorgan. According to the report, the initiative automated commercial loan agreement reviews, saving the company millions of dollars and thousands of hours of human work.
However, there was just one problem: JPMorgan's project actually took place in 2017, years before ChatGPT or any other public-facing AI chatbot was launched.
The string of errors, unfortunately, did not end there. One report on EVs cited a study on air quality in Riyadh that could not be traced to any journal, while another contained footnotes that led to broken webpages.
Another EV report claimed on at least three separate occasions that 90% of road accidents are caused by human error. However, the report cited two different studies for two of those claims and provided no attribution at all for the third.
Speaking to the FT, GPTZero policy analyst Paul Esau admitted that the data on road accidents isn't fake but argued that "no human is going to cite the same fact three times in two pages using three different sources."
Esau explained that "the chaotic signposting of source material is symptomatic of AI-generated research," adding that it is particularly ironic that PwC should make such blunders, given that the company markets itself as an authority on how to use AI responsibly and avoid exactly these types of errors.
It is worth noting that PwC rivals EY and KPMG were also caught passing off AI-generated reports as their own and had to retract those papers after they were flagged by GPTZero.
Like its rivals, PwC also acknowledged its mistake and vowed to change its policies to avoid similar faux pas in the future. In a statement, PwC Middle East said it "takes the accuracy of our published research seriously" and is "updating a limited number of supporting citations."