Deloitte to Refund Australia After AI-Generated Errors in Official Report

Deloitte Australia has agreed to partially refund the Australian government after submitting a report riddled with apparent AI-generated errors. The 237-page report, commissioned by the Department of Employment and Workplace Relations, was found to contain fabricated references, a false judicial quote, and citations of non-existent academic works.

The controversy has sparked a broader debate about the growing use of generative AI tools in professional and government research, especially in areas requiring legal and factual precision. 

Deloitte has since revised the report and confirmed the involvement of Microsoft’s Azure OpenAI system during its preparation.

Key Takeaways

• Deloitte will refund part of the AU$440,000 ($290,000 USD) paid for the flawed report.
• The report included fabricated quotes and references likely generated by AI.
• A revised version now discloses the use of Azure OpenAI in drafting.
• Critics argue Deloitte’s conduct shows “inappropriate misuse” of AI tools.
• The case raises questions about accountability and quality control in AI-assisted research.

The Background: A Report Gone Wrong

The original Deloitte report was published in July 2025 as part of a government review into welfare system automation. It analyzed the Department’s IT systems and their use of automated penalties, an issue that has drawn public scrutiny since Australia’s “Robodebt” scandal.

However, after publication, Sydney University researcher Chris Rudge noticed inconsistencies. He discovered numerous fabricated references and an entirely false quote attributed to a federal court judge. Alarmed by the findings, Rudge went public, calling the errors “hallucinated by AI” and raising concerns about the reliability of AI-generated content in government decision-making.

Read Also: What is AI Mode and How to Use It

What Deloitte and the Department Said

After being alerted, the Department of Employment and Workplace Relations reviewed the report and confirmed that “some footnotes and references were incorrect.” Deloitte subsequently agreed to repay the final installment of its contract.

While Deloitte did not confirm whether the errors were produced by AI, the revised version of the report explicitly disclosed that a generative AI system, Azure OpenAI, had been used in its preparation. The firm emphasized that the “substance” of the report remained intact, and its recommendations were unchanged.

How the Errors Were Detected

Rudge identified up to 20 errors, including one that falsely claimed Professor Lisa Burton Crawford authored a non-existent book. The title, which appeared outside her area of expertise, was immediately suspicious.

“I knew it was either hallucinated by AI or the world’s best kept secret because I’d never heard of the book and it sounded preposterous,” he said.

He also found that several academics’ names were used as citations to give the report a false sense of credibility, even though their actual work had not been reviewed or referenced correctly.

The most serious issue, however, was the misquotation of a judge in a way that could misstate legal principles. According to Rudge, “That’s about misstating the law to the Australian government in a report they rely on. So I thought it was important to stand up for diligence.”

The Role of Generative AI in the Report

The revised report disclosed that Deloitte had used Azure OpenAI, a Microsoft cloud-based large language model platform, to assist in drafting. While generative AI tools are becoming increasingly common for summarizing or analyzing large datasets, they are also known to “hallucinate”, a term for when the model fabricates information that sounds plausible but is entirely false.

Such hallucinations can go unnoticed without human verification, especially in large documents. In Deloitte’s case, the apparent failure to cross-check references and quotes created an embarrassing situation for both the firm and the government.

Political and Academic Reactions

The revelation prompted strong criticism from academics and politicians alike. Senator Barbara Pocock of the Australian Greens said Deloitte had “misused AI and used it very inappropriately.” 

She argued that the firm should refund the entire AU$440,000, calling the mistakes “the kinds of things that a first-year university student would be in deep trouble for.”

Other observers noted that while AI tools can speed up document drafting, their use in sensitive or legal contexts demands stricter oversight. The case highlights the need for transparency in disclosing AI-assisted content and for clear accountability when such systems produce false information.

Broader Implications for AI Use in Government

This case adds to a growing list of incidents where AI-generated material has slipped into official or professional outputs unchecked. In government settings, such mistakes carry higher stakes since reports often inform policy and regulatory decisions.

The Australian government, like many others, is increasingly using AI for research, automation, and analysis. But the Deloitte incident underscores the risks of over-reliance on AI without rigorous human validation. 

It raises urgent questions about how consultants and contractors should disclose AI use in official documents, and what safeguards must be in place to prevent hallucinated content from being treated as fact.

The Line Between Efficiency and Ethics

AI can save significant time in producing drafts, summaries, or literature reviews. However, as this case shows, shortcuts in verification can undermine credibility and damage public trust.

Experts suggest implementing hybrid workflows that combine AI’s efficiency with strict human auditing. Every AI-generated citation or reference should be independently verified, and the use of such systems must be disclosed from the outset.

For global consultancies like Deloitte, the incident serves as a warning that the use of generative AI in high-stakes contexts must be governed by ethical standards, not convenience.

Final Thoughts

Deloitte’s partial refund marks a rare instance where a major firm publicly faced consequences for AI-related errors. While the financial repayment may seem minor compared to the reputational damage, it sets a precedent for how governments and corporations might handle similar cases in the future.

The incident demonstrates the growing tension between innovation and integrity. As AI becomes embedded in professional workflows, transparency and accountability will be the deciding factors between progress and malpractice.

Read Also: Australia Retirement System Embraces Crypto: $4.3 Trillion

FAQ

What did Deloitte refund the Australian government for?

Deloitte refunded part of the AU$440,000 it was paid for a report that contained fabricated references and quotes likely produced by AI.

What AI system was used to write the report?

The revised report confirmed that Microsoft’s Azure OpenAI system was used in drafting the original document.

What kind of errors were found?

Errors included fake book titles, non-existent academic papers, and a fabricated quotation from a federal court judge.

Who discovered the errors?

Chris Rudge, a researcher from the University of Sydney, identified the issues and alerted the media.

How did Deloitte respond?

Deloitte stated that the matter had been resolved directly with the client and that the revised report maintains its original substance and recommendations.

What lessons does this incident highlight?

It underscores the need for transparency, verification, and accountability when using AI systems in producing official or legal documents.