Deloitte refunded the Australian government after an AI-assisted report cited fake sources

A A$440,000 report Deloitte submitted to the Australian Department of Employment included fake academic sources and a fabricated quote from a federal court judgment. Deloitte refunded part of the contract.

Deloitte Australia · Incident Jul 4, 2025 · Indexed May 13, 2026 · 2 sources

When the deliverable is a report, the AI hallucination IS the deliverable. The refund is a price tag on trust.
What
A A$440,000 report Deloitte submitted to the Australian Department of Employment included fake academic sources and a fabricated quote from a federal court judgment.
Incident date
Jul 4, 2025
Who
Deloitte Australia
Failure mode
Hallucination
AI surface
Copilot
Severity
High

What happened

In October 2025, a Sydney academic reviewing a A$440,000 report Deloitte had submitted to the Australian Department of Employment and Workplace Relations discovered non-existent academic citations and a fabricated quote attributed to a federal court judgment. Deloitte issued a revised report and refunded part of the contract.

The case is the most cited example of professional-services AI failure at consulting scale. It also illustrates an underappreciated dynamic: when the deliverable is a report, the hallucination is the deliverable. There is no downstream system that catches the error.

What broke inside the model

Failure path · mode profile · Hallucination
  1. 01 · TriggerA user asks for a fact, a citation, or a figure.
  2. 02 · Model stepThe model writes a fluent, confident answer.
  3. 03 · Control gapNothing ties the claim back to a real source.
  4. 04 · FailureA fabricated fact ships as if it were verified.
  5. 05 · ConsequenceThe false claim reaches a customer, a court, or the public.

Confidence holds, and even spikes, as the claim detaches from any source.

The model generated plausible-sounding citations that did not exist. The mechanism is hallucination amplified by the absence of a verification step inside the consulting team's workflow. The model produced output. The team accepted the output. The output went to the client.

Public visibilityHigh
Regulatory exposurePossible
Customer impactFew customers
Financial impactDisclosed
Time to disclosureMonths

Partial A$440,000 refund, reputational

  1. PressDeloitte refunds Australian government over AI in reporttheregister.com
  2. PressDeloitte to refund Australian government after AI hallucinations found in $440K reporttheguardian.com
Permalinkhttps://failureindex.ai/failures/deloitte-australia-government-report-hallucinations
CitationAI Failure Index. "Deloitte refunded the Australian government after an AI-assisted report cited fake sources" (FI-0017). Realm Labs. https://failureindex.ai/failures/deloitte-australia-government-report-hallucinations (indexed May 13, 2026).
Share cardA branded image of this record for posts and slides.

Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0017. Full dataset at /data.

Note from Realm Labs, the Index steward

How Realm would make this catchable

Controls for this failure mode
  • Prism
  • OmniGuard
  • AI Detection & Response (AIDR)

For a deployment producing client-facing reports, Realm reads each citation against an authoritative source before the report renders. Unverifiable citations are flagged at draft time. The mechanism is the same one that catches legal hallucinations, applied to consulting workflows.