'Human intelligence problem': Labor senator slams Deloitte's AI bungle (www.afr.com)

Australia’s Labor senator Deborah O’Neill has publicly demanded that the Albanese government seek a full refund from Deloitte Australia after a A$440,000 departmental report needed revision due to errors the firm attributes to the use of artificial intelligence. Deloitte has acknowledged AI involvement and will provide an unspecified partial refund to the Department of Employment and Workplace Relations, but O’Neill framed the episode as a “human intelligence problem,” arguing inadequate oversight of AI-generated content produced misleading or incorrect material in a government-commissioned study. The episode is a cautionary moment for the AI/ML community and public-sector procurement: it highlights real risks when large language models (LLMs) or other generative tools are used in analyst outputs without robust provenance, verification, and human-in-the-loop validation. Typical failure modes—hallucinated facts, misattributed sources, out-of-date training data and brittle prompt engineering—can compound into policy-relevant errors if unchecked. The incident strengthens calls for contractual transparency about AI use, formal QA and model-audit requirements, reproducible citations, and clearer liability rules for consultancies relying on generative models. For practitioners and buyers alike, the takeaway is that tooling must be matched by governance to prevent costly, reputation-damaging mistakes.