Deloitte agrees to refund Australian government after AI hallucinations found in report

Accounting firm Deloitte has agreed to partially refund the Australian government after it was essentially proven that its report contains instances of content generated by AI.

AI promo image — *Image via Tara Winstead (Pexels)*

Deloitte is a pretty well-known UK-based company that is also a member of the "Big Four" accounting firms, with the others being EY, KPMG, and PwC. However, despite its size and reputation, it seems that even Deloitte is not immune to using AI tools to generate commissioned reports, though poorly.

The Register reports that Deloitte was commissioned by Australia's Department of Employment and Workplace Relations (DEWR) last year, with the work revolving around the Targeted Compliance Framework. This is the government's IT system to track and penalize welfare recipients who miss certain mandates, such as job interviews.

Although the report was published in July, many were quick to find fabrications in the work, such as fake citations, imaginary quotes, and non-existent footnotes. It was theorized that this was the output of an AI tool, given the nature of the "hallucinations".

Deloitte began its investigation into these mistakes, and it seems like it has admitted that it used AI for at least a portion of its work. The latest version of the report, published a couple of days ago, contains a disclaimer that Deloitte leveraged Azure OpenAI GPT-4o hosted on DEWR's tenant account to fill certain "traceability and documentation gaps".

As a result, Deloitte has agreed to partially refund the DEWR's agreement, which was worth AUD 440,000 (USD ~290,000). An anonymous source told media outlets that the accounting firm had blamed human error for the mistakes in an internal review rather than admitting poor quality AI text generation.

Although this case did result in Deloitte admitting at least some form of wrongdoing, it does set a worrying precedent where it seems like even a Big Four firm being paid hundreds of thousands of dollars can't be bothered with original research. It also sounds the alarms for executives who spearhead pivots in strategies or make major financial decisions for their companies based on AI-generated reports produced by consultancy firms.