Consulting firms have spent years positioning themselves as guides through artificial intelligence adoption. They produce reports. They advise boards. They charge millions. Yet several now find themselves caught in the very trap they warn clients about.
KPMG released a report last October titled “Total Experience: Redefining Excellence in the Age of Agentic AI.” It painted a picture of businesses racing ahead with autonomous AI agents. UBS supposedly integrated them across investment advisory, risk management and compliance. Swiss Federal Railways deployed agents to plan, book and optimize journeys. Transport for London used them to predict congestion and personalize services. NHS Greater Manchester applied the technology for patient triage and readmission forecasts.
None of it held up.
Spokespeople for the organizations told investigators the claims were factually incorrect, not accurate or misleading. One NHS reference twisted a press release about a lung cancer detection tool into something completely different. Futurism broke the story on June 14, 2026. The publication noted KPMG pulled the document after being alerted. But not before industry outlets and a major Czech newspaper had cited it.
The discovery came from GPTZero. The AI detection firm examined the report and found only five of 45 citations matched their sources. The rest were mangled, misleading, partially fabricated or impossible to verify. “These factual errors are not confined to the report’s footnoted passages,” GPTZero said in its analysis. One example stood out. The document claimed Emirates airline adopted a mobile chatbot named Sara that could converse with passengers and change their flights. In reality Sara is a 2023 robot assistant with no such capabilities.
KPMG responded with a statement. “KPMG International takes the accuracy and integrity of its published content seriously. The report has been removed and we are reviewing the circumstances surrounding its publication. We expect all our people to follow our guidelines on the responsible use of AI, including human oversight to validate content and verify independent sources.” The firm also contradicted its own earlier CEO Outlook survey, claiming 55 percent of chief executives ranked AI as a top priority when the actual figure was 71 percent.
This was no isolated case. Just weeks earlier Australian Financial Review reported that EY withdrew a study on loyalty rewards programs. Researchers at GPTZero found 16 of 27 references hallucinated. The document, which EY Canada consultants used to market cybersecurity services, included made-up data, misattributed citations and a reference to a nonexistent McKinsey report claiming $200 million in unredeemed loyalty rewards globally. The firm took the study offline. It offered no public comment at the time.
Deloitte has faced similar scrutiny. In 2025 the firm produced a 237-page workforce trends report for the Australian government that contained more than 20 fake citations to nonexistent academic papers. It also included a fabricated court quote. Deloitte issued a partial refund of AU$440,000 while insisting the core recommendations remained sound. An Australian senator suggested procurers might be better off with a ChatGPT subscription. Another Deloitte report for the Canadian government on healthcare reportedly carried a $1.6 million price tag and similar fabricated references. The firm refunded part of that contract too.
But why do these errors keep appearing in the work of firms that sell AI governance?
Large language models generate text by predicting the next word based on patterns in training data. When they lack specific information they improvise with plausible-sounding fabrications. The more confident the tone, the harder the mistakes are to spot. Consultants under pressure to deliver quickly turn to these tools for drafting, research and summarization. Human review often proves superficial. The polished output looks authoritative. Boards and clients accept it.
McKinsey’s own surveys show the trend. Its 2025 State of AI report found 88 percent of organizations using AI in at least one business function, up from 78 percent the year before. Inaccuracy ranks as the top reported risk. Nearly one-third of respondents noted negative consequences from AI errors. Adoption races ahead of safeguards.
Edward Tian, GPTZero’s CEO, warned of second-hand hallucinations. Fabrications enter the information supply chain. Other models or reports cite them. The well gets poisoned. This mirrors problems in law where AI tools have generated fake case citations that attorneys submitted to courts.
Consulting firms face particular exposure. Their product is credibility. A single high-profile mistake damages trust across the industry. Clients pay premiums for independent verification and rigorous analysis. When that analysis turns out to rest on invented examples the value proposition collapses.
Some firms now experiment with retrieval-augmented generation systems that ground outputs in verified internal documents. Others add mandatory multi-stage human checks. A few have built custom models trained only on proprietary data to reduce fabrication risks. Progress remains uneven.
The KPMG episode carries particular sting. The report celebrated agentic AI, systems that act autonomously. It positioned consulting firms as leaders in this shift. Instead it demonstrated how easily those same firms can ship unverified machine output as expert insight.
Industry observers note the pattern. Reports that hype AI benefits often rely on the technology itself for speed. The incentive structure rewards volume over verification. Marketing teams push glossy PDFs. Partners sign off without deep scrutiny. The cycle repeats.
Recent coverage highlights the breadth. The Register called the KPMG document an accidental demo of the very problem consulting firms caution clients against. It detailed how the errors extended beyond footnotes into core claims. Sherwood News and LinkedIn discussions amplified the story, noting EY’s study vanished quietly after similar findings.
Financial consequences vary. Partial refunds signal accountability but rarely match the reputational cost. Government clients grow wary. Private sector buyers ask tougher questions about methodology. Procurement teams now demand proof of human oversight and source validation.
The incidents reveal a deeper tension. Consulting giants laid off staff while investing heavily in AI. McKinsey reportedly deployed 12,000 AI agents internally after earlier cuts. The technology promises efficiency. It also introduces new failure modes that traditional quality controls struggle to catch.
Leaders inside these firms acknowledge the challenge. Guidelines exist. Training programs stress verification. Yet competitive pressure and tight deadlines erode discipline. A draft generated in minutes looks complete. The temptation to ship proves strong.
So what comes next?
Firms that treat verification as a billable skill may gain advantage. Those that treat it as an afterthought risk more public embarrassments. Clients will likely demand contractual clauses requiring disclosure of AI use and independent audits of outputs. Insurers may adjust professional liability coverage to account for hallucination risks.
The irony lingers. Reports meant to build confidence in AI instead exposed its limits in the hands of its biggest promoters. Accuracy still depends on people willing to slow down, check facts and stand behind every claim. No algorithm replaces that responsibility. Not yet. Perhaps not ever.
WebProNews is an iEntry Publication