Deloitte Canada is under heavy scrutiny after a government-commissioned healthcare report for Newfoundland and Labrador was found to contain what appear to be AI-generated “hallucinations” in its academic citations.
The episode has become a global reference point in debates on AI governance, public-sector procurement, and professional accountability.
What happened
The 526-page report — focused on healthcare workforce and virtual care and commissioned for just under CAD 1.6 million — was released in May 2025. Investigations by local outlet The Independent uncovered numerous academic citations that could not be located, incorrect attributions, and references that appear entirely fictional. Several named “co-authors” had never collaborated, strongly suggesting that generative AI was used to produce or embellish the citation list.
Deloitte’s explanation
Deloitte Canada maintains that AI did not write the report and that its substantive analysis remains sound. The firm acknowledges, however, that AI tools were used “selectively” to assist with some research citations and that errors will be corrected in a revised version.
Critics argue that the pattern of fabricated references points to far greater reliance on generative AI, and insufficient human verification, than Deloitte admits.
Political and public fallout
The revelations have sparked intense political backlash in Newfoundland and Labrador, where the new Progressive Conservative premier has faced calls to seek a refund, mirroring Australia’s response to an earlier Deloitte AI citation scandal.
The provincial health department has confirmed “irregularities” and asked Deloitte to amend the report, but the original version remains on the government website.
A broader pattern for Deloitte
This incident follows a similar controversy in Australia, where a welfare policy review contained fabricated academic references and even a fictitious judicial quote. Deloitte later disclosed that Azure OpenAI was used and refunded part of its fee to the Australian government. The two cases together raise deeper questions about the use of AI in public-sector work.
Lessons for AI governance in professional services
The Canadian episode is a reminder that generative AI can silently contaminate high-stakes policy advice when it is used as a shortcut for evidence gathering. It underscores the need for transparent disclosure of AI involvement, strict “no unverified citations” rules, and clear consequences for advisory firms that deliver government reports built on invented sources.
For both governments and professional services firms, this has become a case study in why AI governance, human oversight, and quality assurance must be designed in from the outset — not retrofitted once public trust is already damaged.
