Deloitte is facing another demand to repay an Australian government client after a consultancy report prepared with the assistance of generative AI was found to contain incorrect academic references and financial calculations that were out by hundreds of millions of dollars.
The Western Australian government paid Deloitte A$166,000 for a feasibility study examining potential mergers involving Curtin University, the University of Western Australia (UWA) and Murdoch University.
The study suggested that university mergers could generate up to A$7.6 billion in net economic benefits over 20 years. However, errors in the supporting analysis have raised questions about the reliability of the report and the quality controls applied before its publication.
Hundreds of millions in arithmetic errors
The report contained substantial discrepancies in its calculations of merger costs.
In a high-cost scenario involving Curtin and UWA, Deloitte listed costs of A$514 million before the merger’s first day and A$145 million over the following five years. The correct total was A$659 million, but the report recorded A$399 million — an understatement of A$260 million.
Another scenario involving all three universities added A$396 million and A$116 million to produce A$823 million. The correct total was A$512 million, a discrepancy of A$311 million.
The errors, detailed in The Guardian’s examination of the report, appeared in tables intended to support the financial assessment of the proposed mergers.
Other problems extended beyond arithmetic.
The report contained incorrect academic authors, inaccurate publication details and references that did not support the claims attributed to them. One citation linked to an entirely different journal article.
A Grattan Institute study was attributed to the wrong co-author, while another study about international students listed only one of its nine authors.
A World Health Organization reference contained a ChatGPT metadata tag in its URL, suggesting the link had been obtained through the chatbot.
The report also repeatedly referred to Curtin University by its former name, Curtin University of Technology, which it stopped using in 2010.
Further concerns emerged over the report’s interpretation of the underlying evidence. According to Times Higher Education, the study repeatedly described one merger option as offering the largest benefit-cost ratio, despite figures in its own summary appearing to contradict that claim.
Deloitte said its AI work had been checked
Deloitte explicitly disclosed using generative AI to support research collation, information synthesis and editorial proofreading.
The report also assured readers that all AI-assisted outputs had undergone a comprehensive quality assurance review by Deloitte professionals to verify their accuracy, relevance and appropriateness.
Yet incorrect references, unsupported claims and basic calculation errors remained in the published document.
Deloitte initially characterised the referencing problems as minor. It has not established that generative AI caused the arithmetic errors, which could have arisen independently of the technology.
Government demands repayment and corrections
The Western Australian government said it was “extremely disappointed” and had written to Deloitte seeking expedited cost recovery.
Officials said they had been advised that the errors did not affect the underlying modelling or findings, but instructed Deloitte to correct and reissue the report.
The substantive economic assumptions have also faced criticism.
Murdoch University vice-chancellor Professor Andrew Deeks challenged the analysis of student completion rates, international enrolments and research funding, arguing that the study relied on unjustified assumptions.
The government has maintained that any university merger will depend on a further business case.
Deloitte’s previous AI-related report failures
The controversy follows two earlier incidents involving Deloitte reports prepared for governments in Australia and Canada.
In October 2025, Deloitte agreed to refund approximately A$98,000 from a A$440,000 Australian federal government contract after its review of the automated welfare compliance system contained nonexistent academic references, distorted legal material and a fabricated quotation attributed to a Federal Court judgment.
The report was revised and republished, with the updated methodology acknowledging the use of generative AI. The department said its substantive recommendations remained unchanged.
A separate Deloitte Canada report, prepared for the government of Newfoundland and Labrador at a reported cost of C$1.6 million, contained citations to academic papers that apparently did not exist, including work falsely attributed to real researchers.
Deloitte acknowledged using AI to assist with some research citations and said it was correcting the references while standing behind the report’s conclusions.
The latest Western Australian controversy is particularly significant because Deloitte explicitly assured its client that AI-assisted outputs had undergone comprehensive professional quality assurance.
The recurring question is whether the Big Four’s review procedures are keeping pace with their increasing use of AI — and whether clients can rely on professional quality assurance to catch errors before reports are delivered.
Want to stay up to date on all things Big Four around the world?
Check out the News section and subscribe to Big4News for weekly deep dives and briefings.
About the author: Claudine Cassar is the founder and editor of Big4News, covering audit, consulting, regulation and governance across Deloitte, PwC, EY and KPMG.