Deloitte will provide a partial refund to the Australian federal government after errors were discovered in a report it produced. In August this year, Australian Financial Review (AFR) reported multiple errors in the report which also included the use of generative artificial intelligence (AI). The report, commissioned by the Department of Employment and Workplace Relations (DEWR), assessed the Targeted Compliance Framework (TCF), part of the IT system that manages welfare and benefits payments.

Errors found in Deloitte report

The original seven-month review, completed around June 2025 for $440,000 AUD, contained multiple mistakes. These included nonexistent academic references and a fabricated quote from a Federal Court judgment, first noted by Australian welfare academic Dr. Christopher Rudge. An updated version of the report, published on DEWR’s website, corrected more than a dozen references and footnotes, rewrote the reference list, and fixed typographical errors.Deloitte confirmed that some footnotes and references were incorrect and has agreed to repay the final installment under its contract. A DEWR spokesperson said the changes do not alter the report’s findings or recommendations.

AI used in Deloitte’s report

The updated report disclosed for the first time that Deloitte’s methodology “included the use of a generative artificial intelligence (AI) large language model (Azure OpenAI GPT – 4o) based tool chain licensed by DEWR and hosted on DEWR’s Azure tenancy.” Deloitte did not link the errors directly to AI and stood by the substantive content of the review.Labor Senator Deborah O’Neill criticized the approach, saying:“Deloitte has a human intelligence problem. This would be laughable if it wasn’t so lamentable. A partial refund looks like a partial apology for substandard work. Anyone looking to contract these firms should be asking exactly who is doing the work they are paying for, and having that expertise and no AI use verified.”Dr. Rudge noted that while the report contained “hallucinations” likely produced by AI, the conclusions generally align with other evidence.