Deloitte to Refund AU$440,000 After AI Errors in Government Report

UPDATE: Deloitte Australia will refund AU$440,000 (approximately $290,000) to the Australian government following the discovery of significant errors in a report that included a fabricated quote from a federal court judgment and references to non-existent academic studies. This urgent development comes as the firm confirmed that the report, initially published in July 2023, was riddled with inaccuracies attributed to the use of a generative AI language system.

The Department of Employment and Workplace Relations released a revised version of the report on Friday, addressing the serious discrepancies raised by Chris Rudge, a researcher at Sydney University. Rudge alerted the media, describing the original report as “full of fabricated references” that undermined its credibility.

In a statement, the department confirmed that Deloitte had acknowledged several incorrect footnotes and references, leading to the decision to repay the final instalment under its contract. Specific details regarding the amount of the refund will be disclosed once the reimbursement is processed.

Deloitte’s response to media inquiries about the AI’s role in generating these errors was notably vague, stating the “matter has been resolved directly with the client.” The potential for generative AI systems to produce erroneous information, known as “hallucination,” has raised concerns about the reliability of AI-generated content.

The problematic report scrutinized the use of automated penalties within Australia’s welfare system. While the department asserted that the core recommendations of the report remained intact, the revised version now includes a disclaimer regarding the use of Azure OpenAI in its creation. Notably, all quotes attributed to a federal court judge have been removed, in addition to references to non-existent documents.

Rudge highlighted up to 20 errors in the initial report, including a claim that Lisa Burton Crawford, a professor of public and constitutional law, authored a fictitious book. “I instantaneously knew it was either hallucinated by AI or the world’s best-kept secret,” Rudge remarked, emphasizing the absurdity of the claim.

The severity of the errors, particularly the misquoting of a judge, raises alarms about the implications for legal compliance within the government. Rudge expressed his concern, stating, “They’ve totally misquoted a court case then made up a quotation from a judge…that’s about misstating the law to the Australian government.”

Further escalating the situation, Senator Barbara Pocock, spokesperson for the Australian Greens party on public sector matters, has called for Deloitte to refund the entire AU$440,000, criticizing the firm for “misusing AI inappropriately.” Pocock pointed out that such mistakes would jeopardize a first-year university student, highlighting the seriousness of Deloitte’s blunders.

This unfolding situation raises critical questions about the efficacy and oversight of AI in professional environments. As the debate on AI ethics continues, stakeholders are urged to scrutinize the reliability of AI-generated content in official reports.

As developments continue, all eyes will be on Deloitte and the Australian government to see how they address the repercussions of this incident and the broader implications for AI usage in public policy research.