A recent healthcare-report commissioned by a Canadian province has triggered sharp criticism of consulting firm Deloitte after the document—supposedly costing about CAD 1.6 million—was found to contain numerous inaccurate citations believed to result from AI-assisted writing. This marks the second known government contract in recent weeks for which Deloitte produced flawed work: only last month the firm partially refunded AUD 290,000 following a similarly problematic report for an Australian agency. Although Deloitte claims the errors were limited to references and do not undermine the reports’ overall conclusions, the incidents have stoked concerns over the firm’s growing reliance on generative AI and its ability (or willingness) to sufficiently verify outputs before delivery.
Sources: Fortune, Cyber News
Key Takeaways
– The flawed Canadian report, funded by taxpayer dollars, contained fabricated or erroneous citations likely generated by AI rather than genuine academic sources.
– The fact that this is the second high-profile government contract in as many months tainted by AI-related errors shakes confidence in Deloitte’s quality-control procedures — especially for critical analyses in healthcare and public policy.
– Despite admitting citation mistakes and issuing partial refunds, Deloitte’s insistence that the core findings remain valid does little to quell skepticism over the wisdom of using AI tools in high-stakes consulting reports without rigorous human oversight.
In-Depth
The recent disclosure that a healthcare report prepared by Deloitte for a Canadian province contains AI-generated errors has cast a stark light on the perils of uncritical reliance on generative software — particularly when public funds and policy decisions are on the line. According to reports, the document produced for the province’s Department of Health and Community Services cost approximately CAD 1.6 million, but upon review it was found to include a number of faulty citations, including references to academic research and data sources that either do not exist or are misrepresented.
This isn’t an isolated lapse. Only weeks earlier, Deloitte partially refunded an AUD 290,000 contract following a separate report for an Australian government agency in which similar issues — fabricated references, incorrect data, and AI-linked hallucinations — were documented. In that case, the errors reportedly stemmed from use of a generative AI system (Azure OpenAI), which supplied content that appeared authoritative but lacked real-world factual grounding.
Deloitte maintains that the core conclusions of both reports remain valid — and promises to correct citations. But for critics, that misses the point. A consulting firm entrusted with public-sector analysis and advisement bears the responsibility of ensuring not only that conclusions are coherent, but that every data point and citation is verifiable. When a significant fraction of a report’s research framework relies on AI-generated material, the risk of error — or worse, deliberate spin — increases dramatically.
More broadly, these back-to-back failures illustrate a systemic problem in large consulting practices embracing AI without transitioning internal workflows to match. Generative models can accelerate drafting — but they are probabilistic, not factual. When left unchecked, their outputs can propagate misinformation under the guise of expert analysis. This doesn’t just jeopardize a firm’s reputation — it undermines public trust in institutions that rely on firms like Deloitte to shape policy, allocate resources, or recommend major structural changes.
For governments working with external consultants, the lesson should be clear: never take an AI-colored report at face value. Every citation, every data table, every claim must be audited — ideally by vetted humans with domain knowledge. Otherwise, what you get might not be a leverageable analysis, but a polished bundle of confident errors.