Full story
Deloitte AI Errors in Government Report
Deloitte Australia agreed to repay the final installment on an Australian government contract after a welfare compliance review it produced was found to contain numerous errors.
“Deloitte Australia will partially refund the Australian government the 440,000 AUD ($290,000) paid for a report containing numerous errors, including fabricated quotes and nonexistent academic references.”
These errors were widely linked by sources to generative AI, including fabricated quotes and nonexistent academic references.
The 237-page review for the Department of Employment and Workplace Relations (DEWR) was published in July and later revised.
The updated version disclosed the use of Microsoft’s Azure OpenAI and removed false material.
Deloitte confirmed some incorrect footnotes and references while the department said the core findings remain unchanged.
Multiple outlets note the refund pertains to an AU$439,000–AU$440,000 contract, often translated as about $290,000.
University of Sydney researcher Dr. Christopher Rudge is credited with surfacing about 20 problems in the report.
Errors in AI-Generated References
The documented flaws include fabricated judicial quotes, nonexistent academic papers, and erroneous or misattributed references.
Rudge identified around 20 issues, including a serious misquotation of a judge and even a fabricated book attributed to a Sydney University professor.
Other reports detail a fake Federal Court quote, misspelled judge’s name, and incorrect citations linked to Australia’s ‘robodebt’ case.
Collectively, these are widely described as AI “hallucinations,” where generative systems invent or distort sources.
Corrections and AI Use in Report
After media scrutiny, the department and Deloitte removed false references and quotes from the report.
“The revised report disclosed the use of the AI tool Azure OpenAI in its creation, and false quotes and references were removed.”
They disclosed the AI tool used and processed a partial refund while keeping the report’s findings intact.
AP and Fortune report that the revised document explicitly notes Azure OpenAI and that Deloitte confirmed incorrect footnotes and references.
TechRadar adds that DEWR removed over a dozen false references and rewrote sections while maintaining the core findings.
Some outlets also report that authorities are considering stricter AI-use rules for future consultancy contracts.
Public Reaction to AI Errors
The political and public backlash was sharp.
Senator Barbara Pocock of the Australian Greens called for a full refund and condemned the misuse of AI, likening the errors to those by a first-year student.
Labor senator Deborah O’Neill criticized the firm’s quality control as a “human intelligence problem.”
Several outlets describe the episode as embarrassing and sparking national outrage, underscoring reputational damage and trust concerns for both Deloitte and government procurement.
AI Use in Consultancy Risks
Beyond the refund, outlets emphasize broader implications such as reputational risk to major consultancies.
“To prevent similar mistakes in the future, Deloitte has committed to implementing more rigorous vetting processes for AI-generated content, including enhanced human review and verification protocols...”
There is a growing need for transparency around the use of artificial intelligence and stronger safeguards and verification processes.
Several sources report that Deloitte has committed to tighter vetting of AI-generated content.
Experts and commentators warn that undisclosed or unchecked AI can undermine the reliability of evidence in official policy work.
Some outlets describe this as one of Australia’s first major cases where undisclosed AI use in consultancy led to real consequences.
This incident may prompt the development of stronger contract rules regarding AI use.