Deloitte refunds Australia after AI-generated report errors
Deloitte agreed to partially refund an Australian government department $440,000 after a report on welfare system automation was found to contain multiple fabricated references and quotes. The firm disclosed in a corrected version that it had used OpenAI's GPT-4o model, which reportedly generated hallucinated citations that were not disclosed in the original July report.
What happened
Deloitte completed a $440,000 report for the Department of Employment and Workplace Relations on the Targeted Compliance Framework, an IT system that automates welfare penalties, in June 2025. The report, published in July 2025, contained multiple errors including nonexistent academic references and a fabricated Federal Court quote.
University of Sydney academic Dr Christopher Rudge identified the errors. In an updated version published in October 2025, Deloitte disclosed it had used Azure OpenAI GPT-4o in its methodology, which was not stated in the original July report. The updated version removed more than a dozen nonexistent references and rewrote the reference list.
Reported impact
Deloitte agreed to repay the final instalment under its contract. A Labor senator accused the firm of having a 'human intelligence problem'.
Index analysis
Dr Rudge described the errors as 'hallucinations' where AI models fill in gaps, misinterpret data, or guess answers. He noted that in the corrected version, multiple fabricated references were replaced with 'seven or eight' fabricated replacements in some cases, suggesting underlying claims were not based on actual sources.
Governance controls
The Department of Employment and Workplace Relations said the reference and footnote corrections did not change the report's substance or overall recommendations.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.