TL;DR

GPTZero published an investigation on 4 November saying three Accenture-linked papers contain 31 hallucinated references among 245. It extends a pattern the firm says it has now found across the Big Four. GPTZero sells the detector it used, so its findings come from an interested party, though it says a person checked each flagged item.

The papers are a November 2025 policy paper on African value chains, written by Accenture’s African arm with IE University academics, and two May 2026 arXiv papers by Accenture researchers, one of them from a lab in Bangalore. The policy paper carries the largest count, at 15 of its 89 references. The other two have seven of 102 and nine of 54.

The errors go beyond sloppy formatting. GPTZero lists a journal that appears not to exist, an organisation it could not trace, and references pointing at the home pages of bodies such as the FAO. One citation in the Bangalore paper, on a malicious chatbot called FraudGPT, linked to a fake GitHub repository holding real malware. It has since been deleted, but anyone who followed the link beforehand may have infected their machine.

Limits of the evidence

GPTZero concedes that its tools do not “definitively prove” a model wrote the documents, describing the results as circumstantial evidence. It also accepts that Accenture did not formally publish the two arXiv papers, while pointing to the authors’ company emails and bylines.

The Financial Times covered the findings. Its article is behind a paywall, but its standfirst says Accenture will correct the report and blames “human error”. We have not read the company’s full response.

A recurring problem for consultancies

GPTZero says it has identified hallucinations in reports from PwC, KPMG, Deloitte and Ernst & Young over the past year; we covered its findings on EY, KPMG and PwC. The investigation also lands weeks after Accenture limited central Claude funding, which GPTZero uses as its framing device.

In our view, UK buyers of consulting research should take the practical lesson. A big-name logo on a report no longer tells a reader that its sources were checked. Anyone quoting consultancy research in a board paper, tender or policy document should open the references that carry the argument. Clients commissioning such work can write citation checks into the contract, because these cases suggest internal review is not reliably catching the problem.