Deloitte refunds Aussie gov after AI fabrications slip into $440K welfare report

Big Four consultancy billed Canberra top dollar, only for investigators to find bits written by a chatbot

Deloitte has agreed to refund part of an Australian government contract after admitting it used generative AI to produce a report riddled with fake citations, phantom footnotes, and even a made-up quote from a Federal Court judgment.

The consulting giant confirmed it would repay the final installment of its AU$440,000 ($291,245) agreement with Australia's Department of Employment and Workplace Relations (DEWR) after the department re-uploaded a corrected version of the report late last week – conveniently timed for the weekend.

The updated version strips out more than a dozen bogus references and footnotes, rewrites text, and fixes assorted typos, although officials insist the "substance" of the report remains intact.

The work, commissioned last December, involved the Targeted Compliance Framework – the government's IT-driven system for penalizing welfare recipients who miss obligations such as job search appointments.

When the report was first published in July, University of Sydney academic Dr Christopher Rudge spotted multiple fabrications, prompting Deloitte to investigate. Rudge initially suggested the errors might be the handiwork of a chatbot prone to "hallucinations," a suspicion Deloitte declined to confirm at the time.

Now the new version [PDF] contains a disclosure in its methodology section: Deloitte used "a generative AI large language model (Azure OpenAI GPT-4o) based tool chain licensed by DEWR and hosted on DEWR's Azure tenancy" to fill "traceability and documentation gaps." In plain English, the bot helped with analysis and cross-referencing. Rudge called it a "confession," arguing the consultancy had admitted to using AI for "a core analytical task" but failed to disclose this up front. 

"You cannot trust the recommendations when the very foundation of the report is built on a flawed, originally undisclosed, and non-expert methodology," he told the Australian Financial Review (AFR).

Deloitte, which has long touted its AI prowess and sells training on "responsible AI" to clients, is now dealing with the embarrassment of having its own report exposed as AI-tainted. Globally, the firm generates more than $70 billion annually, and an increasing portion of that comes from advising on AI ethics, policy, and deployment – not from accidentally quoting nonexistent academics.

Among the deleted material were citations to imaginary reports by University of Sydney law professor Lisa Burton Crawford and Lund University's Björn Regnell, as well as a fictional extract from the robo-debt case Amato v Commonwealth. Even "Justice Davis" – apparently meant to be Justice Jennifer Davies – made a cameo in the original version, along with a fabricated quote from nonexistent paragraphs 25 and 26 of her judgment.

Despite all this, DEWR insists the "substance of the independent review is retained, and there are no changes to the recommendations." 

A source told the AFR that Deloitte's internal review blamed the gaffes on human error rather than AI misuse – though, given the list of hallucinated references, that defense may test the limits of human credulity.

For a Big Four firm that charges a premium for its rigor, the incident may have a few anxious partners quietly checking whether their PowerPoint decks contain phantom professors.

The takeaway for public sector clients is clear: if you're paying a Big Four firm hundreds of thousands for expert analysis, you might want to make sure the expert isn't a chatbot with a taste for creative writing.

Deloitte did not respond to requests for comment. ®