Deloitte admits using AI to produce report, to pay part of $440,000 fee to Albanese government

Deloitte

Deloitte, one of the world’s biggest consulting firms, has admitted to using artificial intelligence in producing a government report and it didn’t go perfectly. The company has agreed to repay part of a $440,000 fee to the Albanese government after errors were found in a report it created for the Department of Employment and Workplace Relations (DEWR).

The report, commissioned in 2024, was meant to review Australia’s welfare compliance framework and its IT system, which automatically issues penalties to job seekers who fail to meet certain requirements. When the report was first published in July, it highlighted significant problems with the system, including defects in the IT platform and a lack of clear connection between the system and the laws it was meant to enforce. It also described the system as being overly focused on assuming participants would not comply.

However, the report quickly ran into trouble. The Australian Financial Review found multiple inaccuracies, including references that didn’t exist and citations that appeared to be completely fabricated. Experts quickly flagged these issues. Dr Christopher Rudge from the University of Sydney identified what are known as AI “hallucinations.” These occur when generative AI fills in gaps with made-up information, misinterprets data, or guesses answers instead of relying on real sources. In this case, the errors included multiple fake references, suggesting that the claims in the report were not properly backed by evidence.

Deloitte later reissued a corrected version of the report. The updated document included an appendix acknowledging that AI tools had been used — specifically, a generative AI model (Azure OpenAI GPT-4o) licensed and hosted by DEWR. Deloitte insisted that the corrections were limited to references and footnotes and did not change the report’s findings or recommendations. “The updates made in no way impact or affect the substantive content, findings and recommendations,” a spokesperson said.

The situation has sparked criticism from politicians. Labor senator Deborah O’Neill called Deloitte’s work a sign of a “human intelligence problem” and suggested that public agencies need to pay closer attention to who actually does the work they’re paying for. She added, somewhat humorously, that agencies might be better off subscribing to ChatGPT than paying large consulting firms for substandard work.

In short, the Deloitte report shows both the promise and the pitfalls of using AI in important government projects. While AI can speed up research and reporting, mistakes like hallucinated references can undermine credibility, even in reports by some of the largest consulting firms in the world. It’s a cautionary tale for governments, companies, and anyone considering relying heavily on AI for work that matters.