The firm received AUD 440,000 (around 231,200 Swiss francs). for the report, which was commissioned by Australia’s Department of Employment and Workplace Relations (DEWR). The paper examined an IT system designed to automate sanctions within the country’s welfare framework.

According to the «Australian Financial Review» (AFR), the original version of the report was published in July. A revised version was quietly uploaded to the ministry's website last Friday, containing more than a dozen deletions of non-existent references and footnotes, an updated reference list, and several typographical corrections.

The revisions came after Christopher Rudge of the University of Sydney flagged multiple inaccuracies and suggested they might stem from AI hallucinations - instances where generative models fabricate information that appears factual.

Trust in Findings Eroded

The updated report now includes a disclaimer noting that generative AI was used in a step described as addressing «gaps in traceability and documentation».

«Deloitte has admitted to using generative AI for a key analytical task but initially failed to disclose it,» Rudges told the AFR, adding that the report's recommendations could no longer be considered reliable.

The ministry, however, stated that «the core of the independent review remains intact and there are no changes to the recommendations». Deloitte reportedly waived the final payment for the report, though no amount was disclosed.

Beyond the financial impact, the incident represents a serious reputational setback for Deloitte — a firm that advises corporations, governments, and institutions worldwide on the responsible use of AI.