Deloitte has agreed to refund the Australian government for a report that contained fabricated citations and AI-generated content, following an internal review that confirmed the firm had used GPT-4o without proper oversight. The issue came to light in August when discrepancies in the report’s references were flagged, revealing that several citations were entirely fictional or misrepresented. The report, commissioned by an Australian government agency, was intended to provide strategic insights on digital infrastructure and innovation. However, investigators discovered that the document included references to non-existent studies, altered data points, and misleading attributions—hallmarks of AI-generated content that had not been properly vetted. In response, Deloitte conducted an internal audit and acknowledged that its team had used GPT-4o during the research and drafting process. The firm admitted the use of generative AI was not properly disclosed to the client and that standard quality control protocols were not followed. As a result, Deloitte has now agreed to fully refund the government for the work. The incident has raised concerns about the reliability of AI-assisted consulting and the need for stricter oversight when using generative tools in high-stakes public sector projects. The Australian government has since emphasized the importance of accountability and transparency in future engagements, particularly when AI is involved. Deloitte has pledged to strengthen its internal guidelines around AI use, including mandatory training for staff, enhanced review processes, and clearer disclosure requirements when AI tools are employed in client deliverables. The firm also confirmed that affected team members are undergoing additional compliance training. This case marks one of the most high-profile examples of AI hallucinations impacting official government work, underscoring the risks of relying on generative models without rigorous verification. As AI adoption grows across industries, the incident serves as a cautionary tale about the need for human oversight, ethical use, and transparency in AI-driven content creation.