Deloitte Australia has agreed to refund part of the $290,000 it received from the government for a report that contained numerous errors attributed to artificial intelligence. The flawed report, which was submitted to the Department of Employment and Workplace Relations, included a false quote from a federal court judgment and referenced nonexistent academic papers.
Deloitte’s Financial Commitment
- Refund Amount: AU$440,000 ($290,000)
- Original Report Published: July (year unspecified)
- Revised Version Released: Recently after the errors were reported
The inaccuracies in the 237-page document were highlighted by Chris Rudge, a researcher from the University of Sydney specializing in health and welfare law. He noted that the report was “full of fabricated references.” Following his alert to the media, Deloitte reviewed the report and confirmed that several footnotes and references were incorrect.
Reported Errors and Concerns
Rudge identified around 20 significant errors in the initial report. Notable mistakes included:
- A fictitious book attributed to Professor Lisa Burton Crawford
- Misquotes from a federal court judge
- References to non-existent research in law and software engineering
Following the revelations, Deloitte’s department stated that while the inaccuracies were serious, the core substance and recommendations of the report remained unchanged. The revised document included a clarification stating that a generative AI language system, Azure OpenAI, was used during the report’s creation.
Reactions to the Report
Rudge expressed concerns over the gravity of misquoting a judge, emphasizing the report’s role as an audit of legal compliance. Senator Barbara Pocock, spokesperson for the Australian Greens party, criticized Deloitte for misusing AI and called for the full refund of the report’s cost. She indicated that such errors should not occur in professional reports, comparing them to mistakes that would lead to serious consequences for a university student.
Deloitte has stated that the matter was resolved directly with the client and did not clarify whether the errors stemmed from AI systems, known for a phenomenon called “hallucination,” where false information may be generated.
