Deloitte Australia has agreed to partially reimburse the federal government following a scandal related to the preparation of a $440,000 report for government agencies that utilized generative artificial intelligence, resulting in numerous errors and fictitious references in the document.
This is reported by Business • Media
Errors in the Report and Consequences for Deloitte
The report from the Department of Employment and Workplace Relations (DEWR), released in July 2025, aimed to analyze IT systems for automating sanctions in the social sector. However, independent experts discovered that a significant portion of the document was generated using generative AI. The report contained a large number of factual inaccuracies, fabricated references, and so-called “hallucinations.”
“A significant portion of the report was written using generative AI. Consequently, the document had numerous errors, ‘hallucinations,’ and fabricated references.”
Following criticism from the academic community, Deloitte conducted an internal investigation and made changes to the updated version of the report: over ten fictitious references were removed, footnotes were corrected, and the bibliography was rewritten. Additionally, the methodology section of the report now includes a direct reference to the use of the “generative AI model toolkit (Azure OpenAI GPT-4o).”
Challenges of Transparency in AI Usage
The situation has raised significant concerns among the professional community in Australia. Christopher Raj, a researcher at the University of Sydney, noted that the errors in the report are a typical outcome of AI work and questioned the reliability of the document’s recommendations due to the lack of transparency regarding the use of technologies.
The Department of Employment and Workplace Relations emphasized that “the essence of the review has been preserved,” and the conclusions and recommendations remain unchanged. At the same time, officials declined to comment on the issue of full reimbursement for the report and further collaboration with Deloitte. It is only known that the company agreed to make the final payment under the contract, the amount of which remains undisclosed.
In addition to quality issues, the report also revealed flaws in the operations of DEWR itself: lack of documentation, errors in processes, and a “punitive compliance model.” This continued the criticism from the Commonwealth Ombudsman regarding the unlawful suspension of hundreds of social payments.
Experts believe that the incident underscores the need for ensuring transparency when using AI in government and corporate research. Meanwhile, Deloitte, which is actively implementing innovative technologies, may face the risk of losing trust due to insufficient openness regarding the application of AI in its own operations.