Deloitte AI-Fabricated Citations in Government Advisory Reports (2025)

Incident Summary

In October 2025, a law professor reviewing a Deloitte Australia report on the future of work — commissioned by the Australian government for approximately $290,000 — identified over 20 fabricated references within the document.^[1]

The fabricated citations included references to non-existent academic papers and a fabricated quote attributed to a federal court judgment. Deloitte subsequently disclosed that it had used Azure OpenAI in preparing the report and refunded the final payment to the Australian government.

In November 2025, a second incident was reported in Canada, where a provincial government report valued at approximately one million dollars was also found to contain AI-fabricated research produced by Deloitte.^[2]

Key Facts

• Fabricated content: Over 20 fabricated references in the Australian report, including non-existent academic papers
• Fabricated legal citation: A quote was falsely attributed to a federal court judgment
• AI tool used: Azure OpenAI, as disclosed by Deloitte
• Contract value: $290,000 (Australia); approximately $1 million (Canada)
• Detection method: Manual review by a law professor
• Financial outcome: Deloitte refunded the final payment on the Australian contract
• Scope: Two separate government contracts in two countries affected

Threat Patterns Involved

Primary: Overreliance & Automation Bias — Deloitte relied on AI-generated content for government advisory reports without adequate verification of citations and references, allowing fabricated academic sources and legal quotes to pass through quality assurance processes

Significance

This incident demonstrates critical risks in the integration of generative AI into professional advisory services:

1. Citation hallucination at scale — Large language models can generate plausible but entirely fabricated academic references, legal citations, and quotes that are difficult to detect without manual verification
2. Quality assurance gaps — Traditional consulting review processes were insufficient to catch AI-generated fabrications, suggesting the need for AI-specific verification workflows
3. Government decision-making risk — Fabricated evidence in policy advisory reports could influence government decisions on workforce planning and regulation
4. Systemic industry concern — Two separate incidents across two countries suggest the problem may be widespread in professional services firms adopting generative AI tools