KPMG has withdrawn its own report on artificial intelligence usage after several organizations reported the document contained fabricated claims [1].
The incident creates a significant contradiction for the firm, as it actively markets an “AI Trust” service to its corporate clients [1]. By selling the ability to verify and trust AI systems while failing to verify its own published research, the company faces a crisis of credibility regarding its core consulting offerings [2].
The report was pulled after major organizations, including the NHS and UBS, said that the claims regarding their specific AI usage were false [4]. These organizations disputed the case studies and success stories presented in the document, which did not reflect their actual operations [5].
KPMG said the errors were due to AI hallucinations during the report's generation process [4]. Hallucinations occur when a large language model perceives patterns or facts that are nonexistent, leading the AI to confidently present false information as truth [3].
This failure suggests that the firm relied on automated tools to produce the report without sufficient human oversight or fact-checking. The resulting document included bogus case studies that misrepresented how high-profile institutions were implementing technology [5].
While the firm continues to offer guidance on AI governance, the retraction serves as a public example of the risks associated with unverified AI output [2]. The company has not provided a timeline for a corrected version of the report, but the withdrawal was immediate following the complaints from the affected organizations [3].
“KPMG markets an “AI Trust” service to clients but withdrew its own AI‑usage report.”
This incident highlights the 'trust gap' in the current AI consulting market. When a firm specializing in AI risk and governance falls victim to the very hallucinations it is paid to mitigate, it underscores that even professional services firms are struggling to implement the rigorous human-in-the-loop verification required for enterprise-grade reporting.
