The Deloitte WA report on merging three Perth universities is now at the centre of a refund demand. Western Australia’s government says it is “extremely disappointed” and wants “expedited” cost recovery of the A$166,000 it paid, after reporting found a A$260 million arithmetic gap and references that look AI-generated.
The story broke through The Guardian’s reporting on 26 October 2026. Here is what is confirmed, what is disputed, and why the case matters beyond Perth.
Key takeaways
- The WA government paid Deloitte A$166,000 for a feasibility study on merging Murdoch, Curtin and the University of Western Australia (UWA).
- One high-cost scenario lists costs that add up to A$659m but states A$399m, a A$260m gap. A second gap of A$311m appears in a low-cost scenario.
- Deloitte says it used generative AI for research collation, information synthesis and editorial proofing, and called earlier errors minor.
- The government says the findings stand, and Deloitte will review and reissue the report.
What did the Deloitte WA report get wrong?
Two kinds of problems surfaced. The first is sums. In the high-cost scenario for a Curtin and UWA merger, the report lists A$514m in costs up to day one and A$145m over the following five years. Those add up to A$659m, yet the report states A$399m. That is the A$260m discrepancy.
The second kind is references. The Australian Financial Review reported errors bearing “the hallmarks of artificial intelligence mistakes”. Per Times Higher Education on 16 October 2026, three citations appeared to name the wrong study authors, two references did not match the studies cited, and one link pointed to an entirely different journal article.
An AI “hallucination” is when a language model produces confident text that is simply false, such as a plausible-looking citation for a paper that does not exist.
Why does the WA government want its money back?
A government buys a consultant’s report to rely on it. If the numbers do not add up, the buyer has paid for something it cannot use as delivered. Seeking reimbursement is the government’s way of saying the product was not what it ordered.
The stakes are bigger than the fee. The report feeds a debate about reshaping state universities. THE reported on 12 October 2026 that the consultant suggested benefits could “pay for itself 16 times over”. A claim like that deserves clean arithmetic.
Do the errors change the conclusion?
That is the disputed part. Deloitte called the referencing errors minor and said they do not affect the findings. The government says the mistakes do not alter the modelling or findings. Minister Tony Buti said the case for the merger, only one part of the wider tertiary reform, remains solid.
Critics disagree. Prof Stephen Bartos, a public-sector governance expert at the University of Canberra, said the errors discredited the report and that it may need reworking by hand. Murdoch’s vice-chancellor, Andrew Deeks, who opposes the merger, attacked the assumptions themselves. He objected that Deloitte treated part-time students who finish after six years as never finishing, and that it assumed weaker universities close 50% of the gap with the best performer, with 30% and 70% as low and high cases. He called those “nice round numbers”.
THE also noted that the report’s two claims about which option has the highest benefit-cost ratio were inconsistent. It gave 15.5 for the three-way merger at best, and 27 for a Murdoch and UWA merger.
How does this compare with Deloitte’s 2025 case?
This is not Deloitte’s first public AI-and-citations problem in Australia. In 2025 it admitted using Azure OpenAI GPT-4o in a report for the federal Department of Employment and Workplace Relations, after fabricated references and a made-up court quote were found. It agreed to repay part of an original contract of about A$440,000, according to TechRadar.
| Detail | 2025 federal report | 2026 WA report |
|---|---|---|
| Client | Federal employment department | WA government |
| Fee | About A$440,000 | A$166,000 |
| Problems | Fake references, invented court quote | Wrong citations, A$260m and A$311m sum gaps |
| Outcome | Partial refund | Refund sought, report to be reissued |
What is the lesson for anyone using AI at work?
Opinion: the lesson is not “never use AI”. It is that someone has to own every number and every citation. Language models write fluent text, not checked text. A total that does not match its parts is something a human reviewer, or a ten-line spreadsheet check, should catch before a client sees it.
We covered a related finding in Vibe Coding Laziness: people who lean on AI tend to check less. A consultancy charging for expertise has even less excuse.
What happens next
Deloitte is expected to review and reissue the report, and the government is pursuing reimbursement. Times Higher Education reported on 22 October 2026 that the consultants were ordered to redo the report. Watch for whether the corrected figures change the benefit-cost ratios, whether the refund is full or partial, and whether the government tightens its AI-disclosure rules for consultants.
FAQ
How much did WA pay Deloitte?
A$166,000 for the feasibility study on merging Murdoch, Curtin and UWA, according to The Guardian.
Did Deloitte admit using AI?
Deloitte said it used generative AI to support research collation, information synthesis and editorial proofing. It had earlier described the errors as minor referencing mistakes.
Is the merger plan dead?
No. The government says the findings are unchanged and the merger case remains solid, though critics say the report may need rebuilding.
Next step: if your team uses AI for reports, add a rule that every figure and citation is checked by a named person before it ships.
Leave a comment