ConstraintCognitive automation2025-10-07
Deloitte agreed to repay the last instalment of a US$290,000 Australian government report after fabricated references and a misattributed court quote were found; it acknowledged using Azure OpenAI
Accountant / Bookkeeperoccupation page →
Event date / reported
2025-10-07
Evidence stage
ConstraintFailure, rollback, regulation or cost is suppressing adoption. Can lower an assessment or widen its uncertainty.
Tasks this bears on
Supervising the automation itself
Checking what the tools produced, catching silent errors, and owning the result when a model got it wrong.
New task✓ Evidence-backed
Where this applies
An assurance review for Australia's Department of Employment and Workplace Relations by a Big Four firm's consulting practice — professional-services output, not bookkeeping. Corrected report issued 26 Sep 2025; the repaid instalment was later put at about A$97,000 (CFO Dive, 21 Oct 2025). Bears on who owns machine-drafted output, not on accounting tasks directly.
What this means
A constraint record that puts a price on the oversight task: a Big Four firm repaid part of a A$440,000 government fee after fabricated references and a misattributed court quote were found in a report drafted with Azure OpenAI. On the deployment layer it shows that when machine-drafted output goes out unchecked, the firm — not the tool — carries the refund and the correction, which is the 'owning the result when a model got it wrong' work the task page describes.
What it does not yet show
This was a consulting assurance review, not bookkeeping, tax or audit work, so it bears on accountants only by analogy about who owns generated output; the fit to this occupation is the weakest on the site and should be read that way. One repaid instalment (later put at about A$97,000) is one contract, and the record shows nothing about how often generated citations slip through or whether the firm changed its review process afterwards.
What you can check
Take the last report, memo or set of workpapers you issued that contained a reference, citation or quoted figure, and check each one back to its source in under an hour; the number you cannot trace is the exposure this record describes, measured in your own files rather than Deloitte's.
Does it change the assessment?
No. The impact index is never moved by a single event. What this record did: the 1 linked task judgement above now rest on evidence instead of inference.
Source
Fortune · verified 2026-09-10 · Claude (VOLO agent) — source text fetched and cross-checked · interpreted 2026-09-10 · Claude (VOLO agent)