The Western Australian state government is pursuing cost recovery from consulting firm Deloitte after discovering severe accounting errors and AI hallucinations in a $166,000 feasibility study regarding a proposed university merger. The report contained multi-hundred-million-dollar calculation errors in its cost appendices, alongside dead links containing ChatGPT metadata tags and outdated institutional names.

Deloitte acknowledged using generative AI tools for research synthesis, proofing, and data collation during the report’s creation, initially dismissing the issues as minor editorial errors. However, public administration experts noted that discrepancies—such as miscalculating a $659 million sum as $399 million—undermine the validity of the study, prompting state officials to demand a manual review and re-issuance of the document.

The incident highlights the operational and reputational risks associated with unverified generative AI usage in corporate and government advisory work. Enterprise clients are increasingly scrutinizing AI-assisted deliverables for hallucinations and technical inaccuracies.

Why it matters

  • Illustrates the severe financial and reputational liabilities consultancies face when deploying unverified generative AI in client deliverables.

  • Emphasizes the necessity of robust human-in-the-loop audit processes for corporate AI adoption.

  • Signals tightening client oversight and potential clawback risks for service providers using automated drafting tools.

Source: theguardian.com