Deloitte has agreed to refund part of a $440,000 fee to the Australian government after a compliance review report contained multiple errors traced to generative AI use. The incident has sparked pointed commentary from lawmakers, with Senator Deborah O’Neill declaring: “Deloitte has a human intelligence problem.”

The Department of Employment and Workplace Relations commissioned the review in December 2024 to assess its welfare compliance framework. Whilst the report’s substantive findings remain valid, it contained fabricated references, nonexistent academic citations, and invented court case details—classic symptoms of AI hallucination.

When AI Fills Knowledge Gaps with Fiction

University of Sydney academic Dr Christopher Rudge identified the errors, describing them as “hallucinations” where AI models fill gaps, misinterpret data, or guess at answers. The corrected version replaced single fabricated references with multiple citations, suggesting the original claims lacked any specific evidentiary foundation.

Deloitte’s updated report acknowledges using Azure OpenAI GPT-4o for parts of the analysis but maintains the core recommendations remain unchanged. The firm stood by its findings, noting “the substance of the independent review is retained.” However, the damage to credibility illustrates a fundamental principle: AI-generated content without rigorous human verification poses significant reputational and professional risk.

The Human Intelligence Problem

Senator O’Neill’s characterisation—“Deloitte has a human intelligence problem”—cuts to the heart of current AI deployment challenges. Having access to powerful AI tools doesn’t eliminate the need for human expertise; it intensifies it. Organisations must apply critical judgment to determine when AI provides value and when it introduces unacceptable risk.

For UK businesses observing this case, the lesson extends beyond avoiding embarrassing errors. The incident demonstrates that delegating substantive work to AI without appropriate oversight transforms a productivity tool into a liability. Professional services firms, in particular, stake their reputations on accuracy and rigour—qualities that current AI cannot reliably deliver independently.

Implications for British SMEs

UK SMEs considering AI adoption for research, analysis, or content creation should note three critical takeaways from the Deloitte incident:

First, AI tools trained to generate plausible-sounding content will prioritise coherence over accuracy when facing knowledge gaps. Without human verification against primary sources, these gaps become filled with convincing fabrications.

Second, the cost of AI errors scales with the importance of the work. A $440,000 government report receiving public scrutiny faces different consequences than an internal analysis, but the reputational principle holds across contexts.

Third, the phrase “AI-assisted” doesn’t mitigate responsibility for accuracy. Clients and stakeholders expect professional-grade deliverables regardless of the tools used to produce them. Human oversight isn’t optional—it’s the difference between AI as a productivity enhancer and AI as a reputation destroyer.

Finding the Right Balance

Dr Rudge’s measured response provides useful perspective: despite his criticism, he hesitates to call the entire report illegitimate because its conclusions align with other evidence. This suggests AI can play a legitimate supporting role in research and analysis when human experts retain responsibility for verification and interpretation.

The Deloitte case doesn’t argue against AI use in professional services—it argues against abdicating professional judgment to algorithms. For British businesses, the path forward involves deploying AI where it demonstrably adds value whilst maintaining human accountability for deliverable quality.

As Senator O’Neill noted, “Anyone looking to contract these firms should be asking exactly who is doing the work they are paying for.” UK SMEs would do well to apply that same standard to their own AI deployments—understanding that technology amplifies capability only when paired with appropriate human expertise and oversight.

Source: Deloitte to pay money back to Albanese government after using AI in $440,000 report - The Guardian, 6 October 2025