The case and the disputes
According to an analysis by GPTZero cited by IT之家, a report published by Accenture with researchers from IE University in Madrid reportedly contains 15 false citations and entirely fabricated case studies. The document, titled “Leapfrogging Development,” examines how artificial intelligence could increase African agricultural productivity. It is the nature of the cited sources, rather than the topic, that has come under scrutiny.
GPTZero, which released the analysis, argues that fabricated citations pose a concrete threat to researchers, academics, consultants, students and anyone seeking information online: the risk is that responses and summaries generated by chatbots will spread false content and end up “contaminating” information sources. In the available material, the analysis firm did not disclose the method by which it reached its conclusions.
Accenture’s position
Accenture disputes the interpretation of GPTZero’s findings. A spokesperson said that internal checks indicated human activity as the source of the errors and that there is no evidence they stem from artificial intelligence “hallucinations.” The company added that it is working to correct the report and intends to republish it only once corrections are complete.
The case takes on particular significance given the company’s public stance. CEO Julie Sweet warned last year that consultants unable to use artificial intelligence would be “eliminated” as part of the company’s transformation. Last month, moreover, Accenture was forced to introduce new limits on the use of artificial intelligence tools, because employee demand was so high that some monthly bills for tools such as Claude exceeded $1,000.
A problem affecting the entire sector
Accenture’s case is not isolated. According to the material, Deloitte, EY, KPMG and PwC have also been accused in the past of AI-related errors in their reports. The phenomenon has also affected the legal world: the British law firm Pinsent Masons admitted that a document filed with the English High Court contained content generated by artificial intelligence “hallucinations,” and the U.S. law firm Sullivan & Cromwell acknowledged similar errors in filings submitted to a New York court.
Several points remain open, however. Based on the available information, it is not known which citations or case studies are specifically disputed, nor which internal controls Accenture applied before publication. It is also unclear whether GPTZero has made its report public in full. The corrected version of the document has not yet been released: it will be the one that makes it possible to verify the actual extent of the changes and the reliability of the sources.
