Deloitte to refund Australian government for AI-hallucinated report
Deloitte Australia has agreed to partially refund the government for a $440,000 report riddled with fabricated citations generated by AI. The firm quietly admitted using GPT-4o after errors were discovered in the document reviewing welfare compliance systems. Critics question the report's reliability despite official assurances that its core findings remain intact.
In July 2025, Deloitte Australia finalized its "Targeted Compliance Framework Assurance Review," a 273-page document commissioned by the Department of Employment and Workplace Relations (DEWR) to evaluate the government's automated penalty system for welfare compliance. Costing taxpayers nearly $440,000 AUD (about $290,000 USD), the report was published in August.
Shortly after publication, Chris Rudge, deputy director of health law at Sydney University, identified multiple nonexistent citations, including fabricated references to papers by University of Sydney professor Lisa Burton Crawford. "It is concerning to see research attributed to me in this way," Crawford told the Australian Financial Review in August. "I would like to see an explanation from Deloitte as to how the citations were generated."
An updated version of the report, released on October 3, 2025, addressed "a small number of corrections to references and footnotes." On page 58, Deloitte disclosed using "a generative AI large language model (Azure OpenAI GPT-4o) based tool chain" to assess whether system code aligns with business requirements and compliance needs. The original report cited 141 sources, but the update reduced this to 127, removing fake references to Crawford's work and a fabricated quote from federal justice Jennifer Davies (originally misspelled as "Davis").
Deloitte will repay the final installment of its contract, though the exact amount is unclear. A DEWR spokesperson stated that "the substance of the independent review is retained, and there are no changes to the recommendations." However, Rudge criticized the process, saying, "you cannot trust the recommendations when the very foundation of the report is built on a flawed, originally undisclosed, and non-expert methodology... Deloitte has admitted to using generative AI for a core analytical task; but it failed to disclose this in the first place."
This incident highlights growing concerns over AI use in professional services, particularly undisclosed applications in high-stakes government work.