Deloitte has agreed to partially refund an Australian government department after errors were discovered in a report completed partly using the technology.
The Big Four firm had been contracted to conduct an assurance review of the country's Targeted Compliance Framework (TCF), part of the IT system that administers welfare and benefits payments. The firm completed the seven-month project, which was worth 440,000 Australian dollars, around $290,000, in June.
But the final report, published in July, was found to contain multiple errors, including academic references to people who didn't exist and a made-up quote from a Federal Court judgment, the Australian Financial Review first reported.
The errors were noticed by the Australian welfare academic Chris Rudge.
An updated version of the report was published on the website of the Department of Employment and Workplace Relations on Friday. The new copy deletes more than a dozen nonexistent references and footnotes, rewrites the reference list, and corrects multiple typographic errors, the AFR reported.
In the updated report, Deloitte also disclosed that its 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."
The detail that AI was used was not part of the report published in July, according to AFR.
Deloitte "confirmed some footnotes and references were incorrect" on the review and has agreed to repay the final installment under its contract, a DEWR spokesperson confirmed to Business Insider.
The spokesperson added that the changes did not change the review's substance or overall recommendations on the TCF system.
Deloitte did not immediately respond to a request for comment from Business Insider about whether AI caused the errors.
Have a tip? Contact this reporter via email at pthompson@bjinnox.com or Signal at Polly_Thompson.89. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.
