⁠Big Four AI Crisis: This Time PwC Catches Heat For Fake Citations

PwC has become the latest of the world's Big Four consulting firms to be publicly caught using artificial intelligence to produce research reports riddled with fabricated citations, broken links, and references to studies that appear to not exist at all.

An investigation by AI detection platform GPTZero, reported by the Financial Times, examined four thought leadership reports published by PwC Middle East over the past two years. The reports spanned subjects as varied as corporate strategy for "agentic" AI, government public services, and electric vehicle market forecasts across the Middle East. Across all four, researchers found repeated instances of what are now commonly called AI "hallucinations" confident-sounding claims backed by sources that simply do not hold up.

One report cited a study on air quality in Riyadh that investigators could not trace to any journal or author. Another contained footnotes leading to broken webpages, or articles that offered no evidence for the claims attributed to them. In a particularly striking instance, a report cited a teenage blogger on Medium with roughly 280 followers as the source for a real-world AI success story involving JPMorgan. The initiative referenced actually took place in 2017, years before ChatGPT was even launched.

Investigators also found the same claim about traffic accidents repeated three times across two pages of one report, each time attributed to a different source. In another instance, the URL cited in a cybersecurity report carried the tracking parameter "utm_source=chatgpt.com," suggesting the underlying material had been pulled straight from a ChatGPT search rather than verified independently.

Responding to the findings, PwC Middle East said it "takes the accuracy of our published research seriously," adding that it was updating a limited number of supporting citations across the affected reports. The firm did not, however, offer a detailed explanation of how the fabricated references made it past internal review before publication.

PwC's case is not an isolated one. Earlier this year, KPMG withdrew a global report titled "Redefining Excellence in the Age of Agentic AI" after several companies named in it complained that case studies and achievements attributed to them had been invented. The withdrawn report contained false claims and fabricated case studies generated using AI.

EY Canada also pulled a report on loyalty rewards programmes after researchers flagged AI hallucinations, faulty citations, and references to non-existent studies, and said it was reviewing how the report came to be published. Deloitte, too, faced scrutiny after a healthcare report it prepared for the Canadian government was found to contain multiple factual errors believed to have been AI-generated, according to earlier media reports cited by the Financial Times.

Taken together, the incidents have intensified questions about how the world's largest consulting firms are deploying generative AI internally, even as they advise governments and corporations on AI adoption and governance. The pattern points to a shared gap: reports meant to inform corporate strategy and public policy are being produced without adequate fact-checking and human verification.

The irony has not gone unnoticed. The same firms that build their advisory practices around helping clients adopt AI responsibly are now struggling to ensure that AI-generated content within their own published work is accurate, properly sourced, and independently verified. For an industry whose credibility rests on rigour and trust, the repeated hallucinations across PwC, KPMG, EY, and Deloitte's reports raise a pointed question: if consulting giants cannot catch AI's errors in their own research, how much scrutiny is being applied to the advice they hand out to governments and businesses navigating the same technology.

For now, PwC has said it is correcting the affected citations, but the broader debate over accountability, disclosure, and human oversight in AI-assisted research is unlikely to settle any time soon. Clients and regulators alike may increasingly ask consulting firms to disclose when and how AI tools were used in preparing research meant to shape real-world decisions.

Inputs from India Today