Deloitte Charged the Australian Government $439,000 for a Report That Cited Works That Do Not Exist
What happened
Australian academics reviewing a Deloitte report on welfare compliance discovered something consultants are not supposed to hand to governments: references to sources that do not exist. The report, commissioned by the Department of Employment and Workplace Relations and worth $439,000 in taxpayer money, contained citations that had been deleted and a quote from a Federal Court judgment that was misstated. The errors were caught not by the client, not by Deloitte, but by outside researchers reading the document after it was delivered.
The report was prepared to inform Australian policy on welfare compliance, a domain where the accuracy of legal and academic citations carries real weight. Decisions shaped by misquoted case law or fictitious research have downstream effects on people's access to payments and services. When academics flagged the problems, Deloitte acknowledged that generative AI had been used in preparing the report to a limited degree, a disclosure that arrived after the fact rather than as part of the original deliverable.
Deloitte issued a corrected version and offered a partial refund. That response addressed the visible product but said nothing about the process that produced it. A correction and a reduced invoice do not explain how fabricated citations survived whatever internal review the firm applied before submitting work to a federal department, and they do not establish whether similar errors appeared in other deliverables produced with the same workflow.
The incident fits a wider pattern of professional services firms using generative tools to accelerate research-heavy work without updating their quality gates to match. A hallucinated citation in an academic paper is embarrassing. The same hallucination in a government policy report, paid for with public funds and used to set rules governing welfare recipients, is a different order of problem. The gap between what a tool produces and what a reviewer actually checks is where the failure happened here, and that gap was only visible because outsiders read carefully.
This is precisely what a verification record is designed to close. A provable record of what a system generated, which citations were checked against real sources, and who signed off before submission would have either surfaced the errors internally or created a clear accountability trail when they emerged later. Neither existed here. The correction came because academics noticed, not because the process required it.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.