GPTZero flags citations in Accenture-linked research

GPTZero published its investigation on November 4, 2026, reporting that three Accenture products contained 31 hallucinated sources, fabricated claims and other likely signs of AI generation.[1] It counted 15 flagged citations among 89 in the agriculture policy paper, seven among 102 in ContextEcho, and nine among 54 in Authenticity Debt.[2] GPTZero defines a hallucinated citation as one that does not match a real source or combines details from different sources.[2] The company said it manually checked the citations it flagged.[2]

“Leapfrogging Development” was published in November 2025 and co-produced by Accenture Africa and IE University.[1] It examines how AI could help African economies gain value from agricultural exports, including cocoa, coffee and sugarcane.[1] GPTZero highlighted a claim that predictive AI improved detection of cocoa swollen shoot disease in Colombia and Ecuador to 99.1% accuracy.[2] It said the study cited for that claim concerned African cocoa cultivation, not Latin America.[2] Accenture’s stated methodology relied exclusively on secondary sources and did not include fieldwork.[5]

Accenture attributed the errors identified in GPTZero’s report to human error and said there was no evidence that they were generated through AI hallucination.[4] The company said the report would be corrected before being re-released.[4] Accenture declined to explain how it determined the errors were human-made or why it ruled out AI malfunctions.[3] IE University said its authors were reviewing the sources and data, and would correct confirmed errors and strengthen safeguards.[2] GPTZero co-founder Edward Tian said the errors were overwhelmingly consistent with AI malfunctions that can produce plausible but nonexistent references and fabricated case studies.[3]

GPTZero cautioned that its detection tools do not definitively prove an LLM generated the papers or citations.[1] The two arXiv papers were not official Accenture publications, although their authors were Accenture researchers.[1] GPTZero also said code from the first arXiv paper was stored in Accenture’s official GitHub organization.[1] A reference about FraudGPT linked to a fake GitHub repository that contained real malware; GPTZero said readers who accessed it before its deletion sometime in 2026 may have infected their computers.[1] The IE–Accenture research’s public launch took place at Accenture’s South Africa office in Johannesburg as a T20 South Africa Summit side event.[6]