Deloitte To Refund Australian Government Over AI-Generated Errors In Report

One of the world’s largest accounting and consulting firms, Deloitte, has been required to provide the Australian government with a partial refund after the company issued a report, valued at about $440,000, that contained several AI-generated errors.

According to The Australian Financial Review (AFR), the Department of Employment and Workplace Relations (DEWR) released on Friday an updated version of the report delivered by Deloitte Australia, including multiple corrections. The document, a review of an automated system used by the government, contained multiple mistakes, including fake academic references and non-existent quotes from a Federal Court judgment.

The first version of the report was published in July. Just a few weeks later, Dr Christopher Rudge, an academic from the University of Sydney, identified several errors in the document, suggesting they were hallucinations generated by an AI model. Deloitte launched an internal investigation after the errors were made public.

In the latest update, beyond deleting the mistakes and revising the document, Deloitte confirmed the use of generative AI: “a generative AI large language model (Azure OpenAI GPT-4o)–based tool chain licensed by DEWR and hosted on DEWR’s Azure tenancy.”

“This is no longer a ‘strong hypothesis,’” said Rudge to AFR. “Deloitte has now issued a confession, albeit buried in the methodology section. Deloitte has admitted to using generative AI for a core analytical task, but it failed to disclose this in the first place.”

A spokesperson from DEWR said to AFR that Deloitte had “agreed to repay the final instalment under its contract.” The specific amount was not specified.

The Deloitte case raises concerns, especially considering that the firm advises large companies worldwide on the responsible use of AI.

“The incident is embarrassing for Deloitte as it earns a growing part of its $US70.5 billion ($107 billion) in annual global revenue by providing advice and training clients and executives about AI,” states AFR’’s report. “The firm also boasts about its widespread use of the technology within its global operations, while emphasising the need to always have humans review any output of AI.”

This year, AI errors have also taken a toll on other companies. Replit’s AI agent deleted the company’s database in July, and the U.S. courts have issued multiple warnings to lawyers and law firms about AI use after AI-generated errors were found in court filings.