Only five of 45 citations in KPMG's 'Redefining excellence in the age of agentic AI' report were accurate, a discovery that forced the report's withdrawal. A 90% failure rate reveals a catastrophic breakdown in quality control for AI-generated content within a leading professional services firm, far exceeding minor errors, according to Engadget.
KPMG, a firm founded on accurate, trustworthy insights, released an AI report that fell victim to widespread AI hallucinations and factual inaccuracies. The incident undermines the credibility of all AI-driven insights in professional services, demanding an urgent re-evaluation of how firms integrate AI-assisted work.
Consequently, companies will likely become more cautious about integrating AI into critical research and reporting. Increased caution will lead to increased investment in AI verification tools and human editorial processes. The goal is to prevent similar reputational damage and ensure factual integrity.
KPMG's Swift Retraction and Review
- KPMG has since pulled the report and is reviewing its publication circumstances, according to Engadget.
KPMG's rapid withdrawal and internal investigation confirm the serious reputational and ethical implications of publishing AI-generated content without sufficient human oversight. The firm's response shows it recognizes the factual errors' severity.
The Extent of AI Hallucinations
Organizations like UBS, the UK’s National Health Service, Swiss Federal Railways, and Transport for London publicly denied claims about their AI usage in the report, calling them untrue or misleading, according to TechCrunch. The public repudiations escalate KPMG's internal quality control lapse into a widespread public trust crisis for AI-generated content.
GPTZero investigators found the KPMG report contained numerous AI hallucinations, including fake footnotes and non-existent AI capabilities, according to Engadget. Roughly half of the claims in the paper were fake or misattributed. The pervasive nature of these fabrications, from client misattributions to fake citations, exposes a critical breakdown in KPMG's quality control processes for AI-assisted content.
How the Inaccuracies Came to Light
The report, initially released in October 2025, contained inaccuracies identified by the research group GPTZero, according to IndexBox. While IndexBox states an October 2025 release, other reports imply the report was published and withdrawn in 2026, according to Engadget, indicating a timeline discrepancy in the reporting.
That an external group, GPTZero, uncovered these extensive inaccuracies post-publication suggests KPMG's internal vetting for AI-generated content was either non-existent or critically insufficient. The uncovering of these inaccuracies raises serious questions about their capacity to responsibly advise clients on AI adoption. The identification of these errors by an external AI detection firm like GPTZero demands independent verification in an age of proliferating AI-generated content.
Implications for AI in Professional Services
GPTZero's finding that only five of 45 citations were real proves professional services firms are ill-equipped to integrate AI into client-facing deliverables without severe reputational damage. The finding directly undermines their core value proposition of trusted expertise, turning AI from a tool into a liability if unverified.
Public denials from organizations like UBS and the UK's NHS, reported by TechCrunch, confirm that unchecked AI output can directly erode client trust and force public retractions. The incident mandates that firms establish rigorous, human-centric validation protocols for any AI-generated thought leadership. Failure to do so risks not only reputation but also client retention and future contract opportunities.
By Q4 2027, if professional services firms like KPMG do not implement clear, human-centric validation frameworks for AI-driven insights, they will likely struggle to regain market trust and mitigate significant reputational and financial risks.
