Can AI-assisted reports pass quality checks with fabricated citations?

Deloitte will partially refund Australia's Department of Employment and Workplace Relations (DEWR) $440,000 after a report it was paid to write turned out to contain "inaccurate footnotes and made up references and quotes." University of Sydney academic Chris Rudge found "over a dozen fabricated points" attributed to colleagues whose work he knew personally: "I was in no doubt, when I read the names of the works, that they were fake because I work in the area and I know my colleagues' work." Among the fabrications: a quote attributed to "Justice Davis" in a 2019 Federal Court case (misspelling Justice Davies) citing paragraphs 25 and 26 of consent orders that contain no such paragraphs, and two cited papers — one in the UNSW Law Journal — that do not exist.

Deloitte confirmed it used "a generative artificial intelligence (AI) large language model (Azure OpenAI GPT-4o) based tool chain licensed by DEWR and hosted on DEWR's Azure tenancy" in producing the report, but has not confirmed the fabrications were caused by that tool chain; a spokesperson said only that "the matter has been resolved directly with the client." DEWR's own statement drew a sharp line between the errors and the work's conclusions: "The substance of the independent review is retained, and there are no changes to the recommendations." Rudge's own read inverts the usual framing of AI error as the worse outcome: "I think it's good if it's AI, because to think of a person doing that is almost worse."

The fabricated-citation mechanism matches Do small law firms misuse AI more often than large ones?, where hallucinated case law slipped into court filings past the filer's own check. This case extends that pattern from solo and small-firm litigation to a named Big Four consultancy delivering a six-figure government contract — the same failure mode (invented sources that read as real) surfacing in higher-stakes, better-resourced professional work. It also sits against Does disclosing AI use damage how trustworthy you seem?: Deloitte disclosed the tool chain only after outside detection forced the question, and even then declined to say whether the AI caused the errors, consistent with a disclosure cost that gives firms reason to confirm as little as possible.

The excerpt is one detected incident, not a rate: it does not say how many other DEWR-funded reports used the same tool chain without errors being caught, nor whether Deloitte's internal review process checked citations before publication and failed, or skipped that check entirely. It also does not establish that AI caused the fabrications — Deloitte never confirmed this, and the report's content changes (two silent revisions before the refund) suggest the firm treated citation accuracy as separable from substance. The incident supports, at the strength of a single case, that a known-reputable firm's AI-assisted output can carry fabricated citations past its own quality control into a paid government deliverable; it does not support a claim about how often that happens.