The Puzzle Behind Deloitte and AI Generated Citations

Before Deloitte could even catch its breath from the last AI citation fiasco, the firm is back in the spotlight like it accidentally wandered into the wrong Zoom meeting. Earlier this year, its AI assisted welfare review for the Australian government turned into a 237 page headache packed with fake citations, a fabricated court quote, and a partial refund that made headlines across the globe. We broke down that saga in our earlier piece, How Deloitte’s AI Error Mistake Sparked Global Debate. Now the plot has a Canadian sequel. A 526 page, almost 1.6 million dollar health workforce report for Newfoundland and Labrador is catching heat for allegedly AI flavored citations that do not exist, misattributed authors, and several academics wondering why they have been credited with research they never touched. 

From Aussie Welfare to Atlantic Health Care 

Quick recap of episode one. In Australia, Deloitte was hired for a welfare compliance review, delivered a report that used Azure OpenAI, and eventually admitted that some references and even a court quote were wrong. A revised 237 page report was posted and Deloitte agreed to partially refund a contract worth roughly 290,000 US dollars. 

Episode two trades kangaroos for Atlantic wind chill. Newfoundland and Labrador’s former Liberal government commissioned Deloitte to build a 10 year Health Human Resources Plan, modeling supply and demand for 21 health occupations and 36 physician specialties. The final report, released in May 2025, runs 526 pages and cost about 1.6 million Canadian dollars, paid over eight installments between March 2023 and March 2025. 

The province expected a grounded roadmap to fix chronic shortages of nurses and doctors, plus guidance on virtual care, recruitment incentives, and COVID era staffing strain. What it actually got, according to local investigations, included citations that do not exist, mismatched author lists, and papers no one can seem to find. If you are thinking this sounds a lot like AI hallucinations slipping into high priced policy work, you are not the only one. 

Fake Footnotes, Real Money 

The Independent, a Newfoundland outlet, dug into the 526 page report and discovered at least four citations that simply do not exist. One was supposedly published in the Canadian Journal of Respiratory Therapy, but no trace of it appears in the journal’s database. Several researchers listed as coauthors say they have never worked together, and others say the cited paper credited to them is pure fiction. Health scholar Gail Tomblin Murphy, who appeared on one of these phantom citations, said the errors look like heavy AI involvement and warned that anything used in public planning must be validated and reliable. Deloitte has acknowledged the four incorrect references, agreed to review every citation in the report, and still maintains that the overall conclusions remain solid. 

Here is how the fallout is stacking up: 

• Government reaction: The Department of Health has ordered a full review of the citations and is waiting for Deloitte’s findings before deciding how serious the issue is. 
• Union pushback: Public sector unions are showing zero chill. The nurses’ union warns that sloppy AI use could point staffing plans in the wrong direction during an already tight labor crunch. 
• Accountability concerns: Another union leader noted that if an employee had fabricated citations, there would likely be discipline. Expecting softer treatment for outside consultants makes no sense. 
• Political pressure: Opposition leaders are now raising refund questions, especially since the province already struggled with an Education Accord that cost more than 750,000 Canadian dollars and also relied on non existent citations. 
• Refund still unclear: Whether Newfoundland and Labrador will push for reimbursement the way Australia did is unknown. The government is holding its cards until Deloitte’s review is done.

Why Finance Pros Should Care Big Time 

Maybe you are thinking this is interesting but far removed from your own work. Spoiler: it is not. If a marquee consultancy and a provincial government can end up with AI tinged fiction in a flagship workforce plan, your next strategy report, risk memo, or valuation deck is not magically immune. 

For CFOs, controllers, tax leaders, and internal auditors, this is a real time lesson in: 

• Vendor risk: Are AI tools being used behind the scenes without disclosure? 
• Assurance gaps: Who checks that case law, research, and benchmarks in third party reports actually exist? 
• Accountability: Who is responsible when hallucinated content slides into a sensitive policy or compliance analysis?

For a profession powered by evidence, this should be a heads up moment. 

Taking Your AI Controls Up A Notch 

A few practical moves are basically a no brainer right now. 

• First, insist on AI disclosure in engagement letters. If tools are involved in research or drafting, you should know it. Full stop. 
• Second, treat citations like financial data. Spot check them. If a ruling or academic paper cannot be found, put the brakes on until someone produces it. 
• Third, tighten internal AI policies. AI can speed you up big time, but it cannot become a silent source of record for technical or regulatory claims. 
• Fourth, negotiate clawback clauses. Australia already showed that refunds are fair game when AI errors contaminate deliverables. 
• Fifth, protect brand trust. A single embarrassing AI citation mess can cost credibility faster than you can schedule a triage meeting. 

Conclusion

At the end of the day, this is not just a Deloitte problem or an AI problem. It is a trust problem. Governments, boards, and finance teams are paying real money for reports meant to anchor staffing plans, policy decisions, and public spending. If the footnotes wobble, everything stacked on top starts to wobble too. For accounting, tax, and finance pros, the message is clear. Use AI to speed up the work, but tighten your guardrails so your evidence chain stays clean. Ask uncomfortable questions, demand clarity on tool use, and write AI clauses that actually have teeth. The teams that get this right will not just avoid embarrassment. They will be the ones clients call when everyone else is scrambling to explain why their research cites studies that never existed in the first place. For more sharp analysis and stories that keep finance pros a step ahead, explore MYCPE ONE Insights.