The merger report has a basic arithmetic problem. Deloitte is under pressure after a feasibility study for three Western Australian universities contained AI-hallucinated references, outdated institutional details and cost calculations that diverge by hundreds of millions of dollars.
The Western Australian government paid Deloitte $166,000 to examine a potential merger involving the University of Western Australia, Curtin university and Murdoch university. Deloitte considered three options between the universities, using “low”, “mid” and “high” cost scenarios, with the supporting calculations set out in the report’s appendix.
Under the “high” cost scenario for a Curtin and UWA merger, the report lists “total costs to day 1” at $514m and “post day 1” costs over five years at $145m. Those figures total $659m, yet the report records $399m — a $260m discrepancy that is difficult to file under harmless formatting.
The numbers do not merely wobble
A second example is even larger. Under the “low” cost scenario for a merger between Curtin, UWA and Murdoch, the report gives total costs to day one of $396m and post-day-one costs over five years of $116m.
That calculation produces $512m, but the report calculates the two figures to $823m — a $311m discrepancy. Prof Stephen Bartos, an expert in public sector governance at the University of Canberra, said the scale of the errors discredits the report and raises concern about whether its conclusions can support a major public decision.
“In my view, they’re so extensive that you probably need a new report, or at the very least someone to go through that material from the ground up and rework it manually without the use of AI,” Bartos said. “Any errors in a report like this, especially on something controversial like a university merger, discredits the report, and Deloitte shouldn’t have had these errors.”
His warning is blunt because the underlying decision is consequential. “Unless the numbers add up, a government might not want to do it,” he said.
AI references add a second credibility problem
The cost discrepancies follow the discovery of hallucinated AI references in the same report. One reference to a World Health Organisation report carries a ChatGPT metadata tag on the URL, indicating that the link was directly sourced from the chatbot.
The report also repeatedly calls Curtin university the Curtin University of Technology, a name the institution has not used since 2010 and one changed under an act of state parliament in 2017. The details do not prove that the modelling is wrong, but they make the document look less like a checked business case and more like a draft that escaped review.
Deloitte admits that it used generative AI to “support the collation of research, synthesis of information and editorial proofing”. The firm had earlier dismissed the errors as “minor”, while the state opposition labelled the report “AI slop” after the hallucinated references emerged in the media earlier in October.
The WA tertiary education minister, Tony Buti, initially described the errors as “typos”. A day later, he said he was angry and disappointed with Deloitte — a change in tone that tracks the discovery of the financial discrepancies rather neatly.
A spokesperson for the WA government said it was “extremely disappointed” and seeking “expedited” cost recovery from Deloitte. The Department of Energy and Economic Diversification has written to the consultancy to accelerate that process.
“We are advised they do not affect the modelling in the report or its findings, but any errors will be corrected as part of a review by Deloitte which will re-issue the report,” the spokesperson said. “All available data, along with new information, will be interrogated by the university reform business case.”
That leaves the government defending the report’s conclusions while asking the company that produced it to repair the evidence underneath them. Deloitte will correct and re-issue the report after its review, but the trust problem will not be fixed by replacing a few references and changing a total.
The wider backdrop is already difficult for AI-assisted work. Australia’s climate has warmed by more than 1.5C since 1910, according to the Bureau of Meteorology’s main long-term temperature data, while only 30% of Australians believed last year that AI’s benefits outweighed its risks. In 2026, community trust in AI appears to be falling even further.
A government report built with AI does not need to be perfect. It does need to name the right university, cite real references and add up — a modest standard, but still one technology has yet to automate.
Based on
- ‘Extremely disappointed’: WA government seeks reimbursement from Deloitte over flawed university report — theguardian.com
- The Australian reports ‘widespread errors’ in BoM temperature data. Where is the evidence for those big claims? | Graham Readfearn | The Guardian — theguardian.com
- Labor and the Coalition are poised to take a reckless gamble on AI. Haven’t they realised community support has collapsed? | Edward Santow and Sarah Sacher | The Guardian — theguardian.com
- The review panel: Wormald decision | Editorial complaints and corrections | The Guardian — theguardian.com
