Deloitte to Refund Australia Over Report with AI-Generated Errors

Deloitte Australia has agreed to partially refund the AU$440,000 ($290,000) it received from the Australian government after a report it prepared was found to contain AI-generated errors, including fabricated legal quotes and references to nonexistent academic studies.

The report, commissioned by the Department of Employment and Workplace Relations, was first published in July and reviewed the department’s IT systems and use of automated penalties in welfare services. A revised version of the report was released on Friday, following media attention sparked by Chris Rudge, a researcher at Sydney University, who discovered numerous inaccuracies and “fabricated references.”

AI Involvement and Inaccurate Citations

After conducting a review, Deloitte and the department confirmed multiple incorrect footnotes and citations. In its statement, the department said Deloitte had agreed to repay the final instalment of its contract, with the refund amount to be disclosed once repayment is completed.

The updated report now includes a disclosure that Microsoft’s Azure OpenAI language system was used in drafting parts of the document. It also removed a fabricated quote attributed to a federal court judge and eliminated references to non-existent academic reports supposedly written by legal and software engineering experts.

Researcher Exposes AI Hallucinations

Chris Rudge, who identified up to 20 major errors, said one of the most glaring mistakes was a reference to a nonexistent book allegedly authored by Lisa Burton Crawford, a professor of public and constitutional law at Sydney University.

“I instantly knew it was either hallucinated by AI or the world’s best kept secret because I’d never heard of the book and it sounded preposterous,” Rudge said.

He added that several citations appeared to use real scholars’ names merely as “tokens of legitimacy”—suggesting the material had not been properly reviewed. Rudge noted that misquoting a federal judge in a report effectively auditing legal compliance was especially concerning.

Deloitte’s Response and Broader Implications

In a statement to the Associated Press, Deloitte said the issue had been “resolved directly with the client,” but declined to clarify whether the inaccuracies were produced by AI.

Experts say the incident highlights the risks of AI hallucinations, where generative AI systems fabricate data, quotes, or references that appear credible but are entirely false.

While the substance and recommendations of the report reportedly remain unchanged, the controversy raises broader questions about ethical AI use in government consulting and the need for transparency in AI-assisted research and reporting.