Deloitte's $260M Adding Error Has WA Demanding Money Back

By Daniel Rao

Somewhere inside Deloitte's 219-page case for merging three of Western Australia's universities, two numbers were added together and the answer came out $260 million wrong. Not a rounding error. Not a modelling judgment call. Addition. And nobody — not the consultants who wrote it, not the department that commissioned it, not the minister who defended it — noticed until a stranger on Reddit did the sum.

Now the Western Australian government wants its money back.

The state paid Deloitte $166,000 for a feasibility study on merging the University of Western Australia, Curtin University and Murdoch University. government spokesperson told Guardian Australia the state was "extremely disappointed" and that its Department of Energy and Economic Diversification had written to Deloitte to expedite the cost recovery process. The errors, the spokesperson said, were being found now — after the report was published, and after the government had already ordered it rewritten once.

Photo: Winthrop Hall, the landmark at the heart of the University of Western Australia's Crawley campus — one of the three universities Deloitte's report proposes merging. Credit: gobeirne / Wikimedia Commons, CC BY-SA. (File photo.)

The sums that don't add up

Start with the arithmetic, because it is the part nobody can spin. Under the report's "high" cost scenario for a merger between Curtin and UWA, it calculates "total costs to day 1" as $514 million and "post day 1" costs over five years at $145 million. Add those together and you get $659 million. The report wrote $399 million — a $260 million discrepancy, in the wrong direction, on one of the scenarios at the heart of the merger case.

We checked the arithmetic ourselves: $514 million plus $145 million is $659 million, not the $399 million printed in the report. It is not a matter of interpretation. It is the kind of sum a pocket calculator gets right.

Then there is the second one, in the other direction. Under a "low" cost scenario covering all three universities, day-one costs of $396 million and post-day-one costs of $116 million should total $512 million. The report calculates them to $823 million — a $311 million discrepancy the other way. Between the two scenarios, the report's own appendix disagrees with itself by more than half a billion dollars.

The errors do not stop at arithmetic. Guardian Australia found one reference to a World Health Organisation report carrying a ChatGPT metadata tag on its URL — the fingerprint of a link lifted straight from the chatbot. Elsewhere, the report repeatedly calls Curtin by its old name, the Curtin University of Technology — a name the institution has not used since 2010, and one that was formally changed under an act of state parliament in 2017. A report that cannot get the name of one of the three universities right is not a report that inspires confidence about the rest.

"Minor" errors, major consequences

How this unfolded tells its own story. Critics first flagged made-up references earlier in October, starting with a post on Reddit, according to Times Higher Education. Deloitte called them minor errors, and the government told the firm to fix and reissue the report. Then Guardian Australia went through the appendix and found the sums.

The minister's own account of the affair has whiplashed. Western Australia's tertiary education minister, Tony Buti, first described the problems as "typos" — and then, a day later, said he was angry and disappointed with Deloitte. It is hard to blame him for the second reaction. It is harder to understand the first: a $260 million addition error in a document meant to justify a once-in-a-generation restructuring of a state's universities is not a typo. A typo is a misspelt word. This is a report whose numbers do not add up, literally.

The opposition has been less charitable. Shadow education minister Liam Staltari told parliament: "It took literally five minutes to find that error. Did anyone in the government actually check the full report before it was published?" It is a fair question, and it is the one that should worry Deloitte most — because the answer, on the evidence so far, appears to be no. Opposition leader Basil Zempilas has said the entire cost should be refunded immediately.

Deloitte, for its part, has not answered Guardian Australia's questions about the latest findings. The firm admits in the report itself that it used generative AI to "support the collation of research, synthesis of information and editorial proofing" — the corporate euphemism for letting a chatbot gather the research, summarise it and proofread it. What the report does not explain is who, if anyone, checked the chatbot's work.

Deloitte has done this before

This is not Deloitte's first AI-assisted embarrassment at the Australian taxpayer's expense. In 2025, the firm agreed to partly refund the federal government for a $440,000 report that also contained AI-generated errors. Rival firms KPMG and PwC have faced similar problems with AI-written reports this year, according to Guardian Australia's reporting. The pattern is no longer a one-off; it is an industry habit, and governments keep paying for it.

That context matters, because it punctures the most convenient defence available: that this was a freak slip by one team on one report. When the same failure shows up at the same firm twice, and across the consulting industry in the same year, the problem is not a person. It is a business model that sells AI-assisted speed at premium prices and treats checking as somebody else's job.

Prof Stephen Bartos, an expert in public sector governance at the University of Canberra, told Guardian Australia the errors discredit the report outright. "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," he said. "Unless the numbers add up, a government might not want to do it."

Prof Christian Downie, from the Australian National University's school of regulation and global governance, who has previously examined government submissions containing AI hallucinations, put the structural point plainly: "I think we need much more detail about how [AI is] being used and for what parts of the report writing process. If they're going to use it for data collection and if there's concerns about the data, then that needs to be detailed in the report." Without that transparency, he said, it is hard to have confidence in the report — and governments at every level need to rethink the contracts they sign with consulting firms, "to ensure that the proper incentives and penalties are in place."

A merger fight the report was supposed to settle

None of this is happening in a vacuum. The merger itself is fiercely contested. The WA government calls it a "once-in-a-generation" opportunity; Buti told the ABC: "I want to commit to it if the business case stacks up, and then it's a government decision." More than 300 students and staff marched on Parliament House on October 21 against the plan, according to ABC News reporting, and the universities themselves are split. Murdoch University's vice-chancellor, Prof Andrew Deeks, has been blunt, writing that "all that a university merger will do for WA is cost a lot of time, money and energy, while reducing choices for students and alienating alumni" — and, of the latest revelations: "Another fail for Deloitte."

Curtin University's vice-chancellor, Harlene Hayne, has said the merger could cost jobs and student choice and addresses problems that do not exist. The University of Western Australia has not taken an official position. The business case itself is due to begin in late 2026 and finish in 2027, with legislation needing crossbench support in parliament. In other words, the report at the centre of this row is not a historical document. It is the foundation for a decision that has not been taken yet — which is exactly why its arithmetic matters.

Photo: Limestone arches along one of the older covered walkways at the University of Western Australia. Credit: Emma Farmer / Wikimedia Commons, CC BY-SA. (File photo.)

The question nobody at Deloitte will answer

Here is the judgment this story actually demands, and it is not really about AI. Chatbots hallucinate; that is what they do. The scandal is that a firm charging $166,000 for expert analysis apparently let a chatbot's output reach a government — twice, counting the federal report — without a human being adding two numbers together first. AI did not fail alone here. The humans who were paid to check failed louder, and the humans who were paid to oversee them failed loudest of all.

That is also why the refund demand, while satisfying, is the smallest part of the story. The government says it has been advised the errors do not affect the report's modelling or its findings, and that Deloitte will correct and reissue it. But a report whose appendix cannot add up, whose references include chatbot fingerprints, and which misnames one of its three subjects, does not get fixed by a corrected appendix. It gets fixed by starting over, by hand, as Bartos suggests — or by admitting that the business case it was meant to support now rests on a document nobody trusts.

For the rest of the world, the lesson is shoddy in the most useful way: it is a perfect, compact demonstration of what "AI-assisted" actually means inside a consultancy in 2026. It means a chatbot did the research, a chatbot did the summarising, a chatbot did the proofreading, and the invoice still arrived in full. Governments from Canberra to Washington should read that sentence twice before signing the next one. The next contract needs a clause Deloitte's evidently lacked: show your working, name the machine that did it, and pay the money back when the sums don't add up.

What happens next, in plain English

Will Western Australia get its money back? The government has written to Deloitte seeking an "expedited" cost recovery, and the opposition wants a full refund. How much comes back — and how fast — is now a negotiation between the department and the firm. Deloitte has not responded publicly to the latest findings.

Does this kill the merger? No — not by itself. The government still calls it a once-in-a-generation opportunity, and the business case runs into 2027. But a merger case built on a discredited report is a much harder sell in parliament, where it will need crossbench votes.

Was AI really to blame? Partly, and that is the point. Deloitte says generative AI supported research, synthesis and proofreading. The hallucinations and the ChatGPT-tagged link are classic AI failure modes. But the arithmetic errors are human failures: somebody was paid to check, and nobody did.

Has Deloitte responded to the refund demand? No. As of October 26, Deloitte has not responded publicly to the latest findings or to the government's expedited cost-recovery letter — Guardian Australia, TNW and the relays all still report no response. The firm has acknowledged only the earlier "minor referencing errors" characterisation. This post will be updated the moment Deloitte answers.

The WA government says all available data, along with new information, will be interrogated by the university reform business case. That is the right promise. Whether it survives contact with the same process that produced this report is the thing to watch — and we will be watching the reissued report's appendix first. Addition, after all, is the easy part.

Sources: Guardian Australia (via its October 25 reporting on the refund demand and appendix errors); Times Higher Education (the resubmission order, parliamentary exchanges and the Reddit findings); ABC News Australia (the minister's position and the October 21 protest); the Western Australian government's university reform material. Deloitte did not respond to Guardian Australia's questions; its earlier characterisation of the errors as "minor" and its statement on generative AI use are as reported.

Read more from TechPulse USA: Australia's push to make AI giants prove their safety systems work; Canberra's demand that AI firms report rogue behaviour; Australia's digital duty-of-care plan for algorithms.

Daniel Rao covers AI policy and accountability for TechPulse USA. He has reported on AI regulation in Australia, the EU and the US since 2024, and he still checks other people's arithmetic.