Deloitte Australia: a government report with fabricated citations (cautionary case)
A Deloitte report for Australia's DEWR had up to 20 errors, incl. invented books and a fake judge quote; AI use was disclosed later; final payment refunded.
This is a cautionary case. A researcher found up to 20 errors in a report Deloitte Australia wrote for the Department of Employment and Workplace Relations (DEWR), including academic works that do not exist and a fabricated quote from a Federal Court judge. A revised version added a disclosure that Azure OpenAI was used in drafting, and Deloitte agreed to refund the final installment of the contract.
Summary
| Item | Detail |
|---|---|
| Client | Department of Employment and Workplace Relations (DEWR), Australia |
| Industry | Consulting |
| Use case | Process optimization |
| Function | Process optimization |
| Reported result | Negative: contract worth about AUD 440,000 (about USD 290,000); Deloitte agreed to refund the final installment (amount to be published once paid); reputational damage |
| Moments rating | C-tier trap |
| AI use-case score | 42 |
Why it matters
The errors are typical of AI hallucination. The real failure was publishing citations nobody had checked, and disclosing generative AI use only after the errors surfaced. Any document that cites laws, judgments or research needs every reference traced back to the original.
Case details
Background
DEWR hired Deloitte Australia to carry out an assurance review of the welfare system’s Targeted Compliance Framework and of the department’s IT system that applies automated penalties. The contract was worth about AUD 440,000 (about USD 290,000). The roughly 237-page report was published on the department’s website in July 2025.
What they did
In late August 2025 the Australian Financial Review first reported problems with the report. Chris Rudge, a health and welfare law researcher at the University of Sydney, found that it attributed a non-existent book to Lisa Burton Crawford, a professor of public and constitutional law at the university. He said the report was “full of fabricated references” and identified up to 20 errors in the first version. They included citations to academic papers that do not exist, references to non-existent reports by legal and software engineering experts, and an invented quote from a Federal Court judge. Deloitte then confirmed that some footnotes and references were wrong and issued a revised version dated September 26, 2025. It removed the judge quote and the non-existent reports and added a note that a generative AI language system, Azure OpenAI, had been used in writing it.
Results and data
- On October 7, 2025, DEWR said Deloitte had agreed to refund the final installment of the contract, with the amount to be made public once the refund is complete.
- Deloitte said the revisions “in no way impact or affect the substantive content, findings and recommendations”, did not say whether AI caused the errors, and said the matter “has been resolved directly with the client”.
- Greens senator Barbara Pocock said Deloitte had “misused AI” and should refund the full amount.
What to take from it
- Check every citation: for any external deliverable with legal grounds or academic references, trace each AI-supplied reference back to the source.
- Disclose AI use up front: state the scope of AI use in the contract and the deliverable. Disclosing only in a revision cost Deloitte more in public perception.
- Focus the review checklist: claims like “X wrote book Y” or “Judge Z said…” are the easiest to fabricate and should be mandatory checks.
Limits
- The direct loss here was one contract installment plus reputation. In healthcare, legal work or listed-company disclosures, the same kind of error could cost far more.
- Deloitte did not admit that AI caused the errors. The case shows the risk of using AI without checking citations; it does not show that AI cannot be used in professional reports.
Source
Original source: Fortune · Deloitte AI Australia government report
Additional source: Associated Press report (via Yahoo News)
This is an industry research entry: a publicly documented third-party AI case, not a Moments client case. For projects Moments delivered itself, see client cases.
Related
- AI hallucination — AI hallucination is when a language model produces fluent, confident text that is factually wrong, because it predicts likely words rather than checking facts.