Deloitte Refunds Australian Government After AI Hallucinations Discovered in $440,000 Official Report

Dr Christopher Rudge, the academic who initially identified the errors, reported that the document contained artificial intelligence "hallucinations."

Deloitte, the global consulting powerhouse, has committed to returning a portion of its $440,000 fee to Australia's government after admitting to utilizing generative AI tools in their assessment of the government's "Future Made in Australia" program. The department had engaged the firm in 2024 to evaluate the targeted compliance framework and its associated IT system, which automatically imposes penalties on job seekers failing to meet mutual obligation requirements, according to Guardian.

The July-released report was discovered to contain numerous significant inaccuracies, including academic citations referencing non-existent individuals and a fabricated quote attributed to a Federal Court judgment, as reported by the Australian Financial Review.

By Friday, the Department of Employment and Workplace Relations had published a revised version of the report on its website. This updated document eliminated over a dozen fictitious references and footnotes, revised the reference list, and addressed numerous typographical errors.

Australian welfare academic Dr Christopher Rudge, who first detected these errors, indicated that the report contained AI "hallucinations"—instances where artificial intelligence systems generate false or misleading information by filling knowledge gaps, misinterpreting data, or producing speculative answers.

"Instead of simply replacing individual fake references with legitimate ones, they've removed the hallucinated citations and incorporated five, six, or even seven to eight new ones in their place in the updated version. This suggests that the original claims made in the report's body weren't based on any specific evidential source," he explained.

In response, Deloitte acknowledged utilizing AI but clarified that it was merely employed during early drafting stages, with the final document thoroughly reviewed and refined by human experts. The company maintained that AI usage had no impact on the "substantive content, findings or recommendations" presented in the report. While Deloitte admitted to using generative AI tools, they did not directly attribute the errors in the original document to artificial intelligence.

In the revised version, Deloitte formally disclosed that their research methodology incorporated a generative artificial intelligence tool, specifically identifying it as a large language model (Azure OpenAI GPT-4o).

A Deloitte spokesperson confirmed that "the matter has been resolved directly with the client." The Department subsequently verified that the refund process is underway and indicated that future consultancy contracts might include stricter guidelines regarding the use of AI-generated content.

This incident has triggered broader discussions about the ethical and financial implications of utilizing artificial intelligence in high-value consultancy work. As firms increasingly adopt AI for efficiency and speed, questions arise about the extent of human involvement and whether clients receive appropriate value for their investment.

Notably, Deloitte recently established a partnership with Anthropic to provide nearly 500,000 global employees access to the Claude chatbot, highlighting the growing dependence on AI across professional services.

This case represents one of Australia's first major instances where a private organization has faced consequences for undisclosed artificial intelligence usage in government-commissioned work.