KPMG Pulls AI Report After Widespread Hallucinations and Fabricated Claims Emerge

In a striking development that has sent ripples across the professional services and AI industries, KPMG, one of the "Big Four" accounting and consulting firms, has officially retracted a significant report on AI usage. The report, titled "Total Experience: Redefining Excellence in the Age of Agentic AI," originally published in October 2025, was pulled from its websites after numerous inaccuracies, including AI-generated hallucinations and fake citations, came to light.

This incident serves as a stark reminder of the persistent challenges associated with AI reliability and the absolute necessity of rigorous human oversight, even for organizations that advise clients on the responsible deployment of artificial intelligence. It underscores a growing concern that AI, while powerful, remains an unreliable source of factual information without stringent verification.

The Report: Ambition Meets AI's Flimsy Reality

KPMG's report aimed to showcase the cutting-edge adoption of "agentic AI" – AI systems designed to perform complex, automated tasks – across various global organizations. It promised insights into how businesses were leveraging these advanced AI agents to enhance customer experience, optimize operations, and redefine excellence in a rapidly evolving technological landscape.

The study, which was intended to be a flagship publication, included numerous case studies detailing the alleged successful deployment of AI agents by high-profile entities. These examples covered diverse sectors, from finance and healthcare to public transportation, painting a picture of widespread and sophisticated AI integration.

The Unraveling: GPTZero and the Financial Times Expose the Flaws

The report's credibility began to crumble when the AI detection and research group GPTZero identified a significant number of inaccuracies. GPTZero's investigation revealed that out of 45 citations in the report, only a mere five accurately pointed to real, uncorrupted sources. The remaining citations were either entirely fabricated, heavily paraphrased with incorrect details, or too vague to verify.

GPTZero coined the term "vibe citing" to describe the phenomenon where generative AI tools create plausible-looking but ultimately false references, mixing real sources with invented components. This "vibe citing," along with other AI-generated errors, led to approximately half of the claims in the paper being either fake or misattributed.

The findings by GPTZero were subsequently verified by the Financial Times, which reached out to the organizations cited in KPMG's report. The responses were unequivocal: many of the claims were untrue or misleading.

High-Profile Denials: Companies Refute KPMG's Claims

Several major organizations explicitly denied the claims made about their AI usage in KPMG's report. These included:

• UBS Bank: KPMG's report stated that the global wealth manager integrated AI agents across its investment advisory and risk management systems using a custom platform built with Microsoft. A spokesperson for UBS reportedly told the Financial Times that these claims were "factually incorrect" and demanded their removal.
• UK's National Health Service (NHS) Greater Manchester: The report claimed the health service used AI agents to organize patient data, automate referrals, and predict hospital readmissions. An NHS spokesperson clarified that this "doesn't really align" with reality, and the cited source only described AI use in early lung cancer detection, not the broad applications claimed.
• Swiss Federal Railways (SBB): KPMG wrote that the railway used AI agents to help users seamlessly plan and book journeys based on real-time conditions and carbon footprints. An SBB spokesperson confirmed the claims were "not accurate."
• Transport for London (TfL): The study claimed London's transit system was using AI agents to predict congestion and coordinate city transport, a claim dubbed "misleading" by a TfL spokesperson.
• Emirates: The report claimed Emirates launched a mobile chatbot called Sara that could talk to passengers and alter their flights. However, Sara was a mobile assistant launched in 2023 without AI-powered chatbot capabilities to change bookings.
• Japanese East Japan Railway Company (JR East): The report claimed JR East used agentic AI for customer service, citing a 2019 press release—years before agentic AI became mainstream.
• Austrian electricity provider Verbund: KPMG falsely claimed Verbund used AI agents in households for real-time analytics as part of an "energy-as-a-service ecosystem," despite their investment in a startup using agents for grid optimization.

These denials collectively forced KPMG to acknowledge the severe issues and promptly remove the report from its platforms.

KPMG's Response and Internal Investigation

Following the public revelations and complaints, KPMG International issued a statement acknowledging the gravity of the situation. A spokesperson stated, "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."

The firm also emphasized its internal guidelines on the responsible use of AI, which "expect all our people to follow our guidelines on the responsible use of AI, including human oversight to validate content and verify independent sources." This suggests that KPMG's own employees likely failed to adhere to these crucial verification steps, leading to the publication of the error-riddled report. KPMG has launched an internal investigation to understand how these errors occurred and to reinforce adherence to its AI usage policies.

Broader Implications for AI Adoption and Trust

This incident is not an isolated one, but rather the latest in a series of similar occurrences involving prominent professional services firms. Just a month prior, EY withdrew a report on loyalty rewards programs that also appeared to include fake footnotes and AI hallucinations. Last year, Deloitte was compelled to refund the Australian government after AI-generated content and fabricated quotes were found in a taxpayer-funded report. Even the South African government had to withdraw its entire national AI policy after discovering AI-generated fabrications in its academic citations.

These repeated failures by trusted organizations carry significant implications:

• Erosion of Trust: When reports from major consulting firms, often considered authoritative sources, are found to contain fabricated information, it erodes public and corporate trust in AI-generated content and, by extension, in the firms themselves.
• "Poisoning the Well" of Information: As GPTZero CEO Edward Tian warned, error-riddled papers published by "Big Four" firms can "poison the well of information." These reports are frequently cited in other research, news articles, and even by other AI models like ChatGPT and Google Gemini, leading to a "secondary hallucination" effect where misinformation spreads widely.
• The Human Oversight Gap: The core issue remains the failure of human oversight. While AI tools can significantly accelerate content creation, they cannot replace the critical need for human verification, fact-checking, and ethical judgment. The incident highlights that relying on AI without robust human validation is a recipe for disaster.
• Vulnerability of Consulting Firms: The incident underscores how consulting firms, which are often at the forefront of advising clients on AI risks and governance, are themselves vulnerable to AI-generated errors. This creates an embarrassing paradox, as they caution others about safeguarding against the very issues they are falling prey to.
• Call for Stronger Standards: As the use of generative AI becomes more prevalent, there will likely be increased pressure for stronger disclosure, validation, and verification standards for all organizations publishing AI-assisted content. Regulators and governments evaluating AI governance frameworks may soon make these standards essential.

Moving Forward: The Indispensable Role of Human Accountability

The KPMG incident is a powerful lesson in the current limitations of AI and the enduring importance of human accountability. While AI offers immense potential for efficiency and innovation, its outputs must be treated with skepticism and subjected to rigorous human review. Organizations, particularly those in advisory roles, must implement robust internal processes to ensure that AI-generated content is thoroughly fact-checked and verified against independent sources before publication.

This means investing in training for employees on responsible AI use, establishing clear guidelines for content validation, and fostering a culture where critical thinking and verification are prioritized over speed of content generation. The goal should be to leverage AI as a powerful assistant, not as an autonomous and infallible source of truth.

Conclusion

KPMG's retraction of its AI report due to hallucinations is a significant news story that highlights a critical juncture in AI adoption. It's a clear signal that while AI tools are becoming indispensable, their inherent flaws, such as hallucinations and the generation of fake information, require constant vigilance. For the AI industry and businesses globally, this event reinforces the message: AI is a tool that augments human capabilities, but it does not diminish the need for human expertise, ethical judgment, and, most importantly, human verification. The path to truly responsible and trustworthy AI integration lies in acknowledging these limitations and building processes that prioritize accuracy and integrity above all else.

Frequently Asked Questions

What was the KPMG report about?

The KPMG report, titled "Total Experience: Redefining Excellence in the Age of Agentic AI," was published in October 2025. It aimed to explore and showcase how various global organizations were adopting and benefiting from "agentic AI" systems, which are designed to automate complex tasks.

Why was the KPMG report pulled?

KPMG pulled the report because it was found to contain widespread inaccuracies, including AI-generated hallucinations and fake citations. An investigation by GPTZero, later verified by the Financial Times, revealed that many claims about companies' AI usage were untrue, leading to complaints from organizations like UBS, the NHS, and Swiss Federal Railways.

What are "AI hallucinations"?

"AI hallucinations" refer to instances where an artificial intelligence model, particularly a large language model, generates information that is plausible-sounding but factually incorrect or entirely fabricated. These can include inventing facts, misattributing quotes, or creating fake citations, as seen in the KPMG report.

What does this incident mean for the use of AI in professional services?

This incident, along with similar cases involving other consulting firms like EY and Deloitte, highlights the critical need for robust human oversight and verification processes when using AI tools for research and content creation in professional services. It underscores that while AI can boost efficiency, it cannot replace human accountability and thorough fact-checking to ensure the accuracy and integrity of published content.