Deloitte to refund government, admits using AI in $440k report (www.afr.com)

Deloitte Australia has admitted it used artificial intelligence to help produce a $440,000 report for the federal Department of Workplace Relations and will issue a partial refund after the document was found to contain multiple errors — notably three nonexistent academic references and a fabricated quote attributed to a Federal Court judgment. A revised version, quietly uploaded ahead of a long weekend, removes more than a dozen bogus references, rewrites the bibliography and fixes typographical mistakes, drawing scrutiny over how the work was produced and reviewed. For the AI/ML community this is a cautionary example of the practical risks of deploying large language models (LLMs) in high‑stakes professional services without robust verification. The incident highlights common failure modes — “hallucinated” citations and invented quotations — and underscores the need for human‑in‑the‑loop validation, provenance and citation tracking, model documentation, and contractual clarity about automations. It also raises legal, reputational and procurement implications: vendors and buyers must define acceptable tool use, audit trails and quality‑assurance pipelines (automated fact‑checking, source linking or retrieval‑augmented generation) to prevent similar errors and to ensure accountability when LLMs are used in client deliverables.