Deloitte Australia
Deloitte Australia refunds part of a welfare-system review after citation errors and AI disclosure
Deloitte Australia used a generative AI tool, Azure OpenAI GPT-4o, in an assurance review of a government welfare IT system. After the report was found to contain incorrect citations and a wrong summary of a court case, Deloitte reissued it and the client department says over A$97,000 was refunded from a contract originally valued at about A$440,000.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
about A$440,000
Original value of the Deloitte contract, per a government procurement notice cited by CFO Dive2
over 97,000 Australian dollars
Amount Deloitte refunded, per a spokesperson for the client department (reported 21 October 2025)2
What it took
The problem
Australia's Department of Employment and Workplace Relations (DEWR) hired Deloitte Australia to review the IT system behind the Targeted Compliance Framework, the system that applies penalties in the welfare programme. The work was ordered in December 2024 under a contract originally valued at about A$440,000. After the July 2025 report was published, outsiders found references that did not hold up, including a court case summarised wrongly.
What they built
This was a consulting deliverable, not a product. Deloitte's team assessed code quality, testing, documentation and alignment with legislation. In one step, checking whether the system's code could be mapped to business requirements and compliance needs, the revised report says the team used a generative AI model, Azure OpenAI GPT-4o, in a tool chain licensed by DEWR and run in DEWR's own Azure cloud tenancy.
How they did it
The report describes the AI tool as one input to a larger assessment that also used code review, test analysis, document review and interviews. The July 2025 version of the report does not mention generative AI. The September 2025 version adds the disclosure in its methodology section and a "Report Update" note. The note says the update corrects citations and reference list entries that contained errors, fixes the summary of the Amato proceeding, and improves clarity. It says the update does not change the substantive findings or recommendations.
What went wrong
The report went out with incorrect citations and a wrong court-case summary. Press coverage reported a fabricated judicial quote and references to papers that do not exist, and described the errors as AI-generated. The department said Deloitte confirmed that some footnotes and references were incorrect and that the substance of the review was kept. The department's spokesperson later said Deloitte refunded over A$97,000.
Results
A corrected report was published, the department kept the findings, and part of the fee came back. The episode became a widely cited example of AI risk in professional services. Whether the AI tool caused the faulty citations is not stated in the report itself.
Lessons for your project
Disclose AI use to the client when you use it, not after a problem. The original July 2025 report did not mention it; the AI note appeared only in the September 2025 version.
Check every citation and reference in AI-assisted deliverables before they leave the building, especially quotes from courts and academic papers. Wrong references can cost fees and credibility even when the findings stand.
Write down which step AI touched and which it did not. Deloitte's report places the AI tool in one analysis step (code traceability), yet the public debate was about citations, and the report does not settle that link.
Put AI-use clauses in consulting contracts: disclosure, review duties and what happens to fees if errors are found.
Caveats
- The deploying firm's own account is a methodology note in its revised report. It says a generative AI tool (Azure OpenAI GPT-4o), licensed by the client department and hosted on the department's Azure tenancy, was used in a code-to-requirements step. It does not say the tool produced the citation errors.
- The report update note says citations and reference entries contained errors and that a summary of the Amato court proceeding contained errors. It says the update does not affect the findings and recommendations. Wording about AI causing the errors comes from press coverage, and CFO Dive says Deloitte had not immediately responded to its request for comment.
- The refund figure and contract value come from CFO Dive, citing a department spokesperson and an AusTender notice. Other outlets report different totals and dates, and the refund was described at first as the final instalment.
- The deployed date shown is the date of the original report (4 July 2025). The sources do not say when the AI tool was used.
- The department's website was unreachable when this entry was prepared, so the revised report was read from a web archive copy.
- This is a consultant using AI inside a government client's environment, not a public body running its own deployment.
Sources
- Targeted Compliance Framework: Department of Employment and Workplace Relations, Final Report, 26 September 2025
Deloitte, published by the Department of Employment and Workplace Relations · 2025-09-26 · Filing · dewr.gov.au
- Deloitte refunds over $60K for report with AI errors, Australian government says
CFO Dive · 2025-10-21 · News · cfodive.com
Summarized in our own words from the sources above. Spotted an error? See corrections.