KPMG Report Allegedly Included Hallucinated Citations and Inaccurate Agentic AI Case Studies

In October 2025, KPMG published "Total Experience: Redefining Excellence in the Age of Agentic AI." GPTZero later reported that only five of 45 citations accurately matched their sources, while others contained flawed details. The pattern reportedly suggested an unverified LLM research tool, but the system and extent of AI involvement remain unconfirmed.

Risk classification

• Primary risk domain: 3 Misinformation
• Primary risk subdomain: 3.1 False or misleading information

The AI system generated false citations and fabricated case studies, leading to the dissemination of incorrect information in a major professional report.

Additional risk subdomains

• 7.3 Lack of capability or robustness: The AI tool failed to perform reliably, hallucinating references and facts due to a fundamental lack of capability in factual verification.
• 5.1 Overreliance and unsafe use: KPMG researchers over-relied on the AI tool, publishing its hallucinated outputs in a global report without performing necessary human verification.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The incident was caused by an AI-powered referencing tool unintentionally generating hallucinated citations and fabricated claims while attempting to fulfill a research prompt.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Risk Level 3: Limited Risk. The system is an AI-powered referencing and research tool that generates text and citations, falling under the category of AI-generated content which requires transparency obligations so users are aware of its nature.

AI system and alleged parties

• AI system: AI-powered referencing tool
• AI purpose: Writing Assistant; Technical Text Generation
• Behaviour type: Assistant
• Alleged developer: Large language model developers, Generative AI developers, AI research tool developers
• Alleged deployer: KPMG, Big Four consulting firms
• Alleged harmed parties: UBS, Transport for London, Transit systems, Swiss Federal Railways, Researchers, Policy analysts, NHS Greater Manchester, Healthcare organizations, Financial institutions, Epistemic integrity

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Minor, indirect Minor
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Epistemic

Reported: The report explicitly describes epistemic harm through the generation of fake citations, fabricated case studies, and the subsequent spread of these false claims across the internet.

Directly caused: The AI tool generated 40 fake or paraphrased citation titles out of 45, along with fabricated claims about companies like Verbund, East Japan Railway, and Emirates.

Indirectly caused: These fabricated claims and statistics were recycled by industry publications (CXM, CX Dive, Mi3, and a Czech newspaper) and ingested by LLMs like ChatGPT-5.5 and Gemini 3.5, propagating the misinformation.

Inferred additional harm: It is highly likely that many other consulting reports, academic papers, and industry analyses contain similar undetected vibe citations, leading to a broader erosion of trust in published research and the pollution of web-scale training datasets.

People affected

• Occurrences reported: 1
• People reportedly exposed: 250000

Potential causes

Management

• Prioritizing Speed Over Quality: Publishing reports quickly without checking technical accuracy.
• Lack of Quality Standards: No organizational policy against publishing hallucinated citations.

Technology

• LLM Hallucinations: AI tool generated fake references and paraphrased titles.
• Tool Over-Compliance: AI tool over-complied with requests to find agentic AI examples.
• AI-Generated Content Errors: AI detector showed case studies were likely fully AI-generated.

Data Inputs

• Condensed Citation Styles: Shortened endnote formats made source verification difficult.
• Historical Data Conflation: Citing older sources that predated agentic AI technology.
• Incomplete Reference Data: Lacked real authors, URLs, and standard identifiers.

Human Factors

• Lack of Human Verification: No human at KPMG double-checked citations or sources.
• Overreliance on AI Tools: Authors accepted AI-generated case studies on faith.

Process and Methods

• Deficient Fact-Checking: No validation process for claims before report publication.
• Unverified AI Workflows: Use of generative AI for research without quality control.
• Poor Source Attribution: Inconsistent formatting and paraphrased titles used.

Information quality

• Classification confidence: High
• Reason for confidence: The report from GPTZero provides a highly detailed, empirical analysis of the specific citations and case studies in the KPMG report, comparing them directly to the real sources and demonstrating clear patterns of LLM hallucination.
• Ambiguities identified: The exact proprietary AI tool used by KPMG to generate the citations is not named, though its LLM-based nature is strongly established by the error patterns.
• Alternative interpretations: KPMG defenders might argue the errors are human formatting issues, but the systematic paraphrasing and fabrication of components strongly point to AI generation.

Global consulting firm KPMG published a report containing 40 hallucinated citations generated by an unverified AI research tool. While this led to the spread of misinformation across industry publications and LLM training data, the incident represents a corporate quality control failure rather than a threat to national security, carrying negligible impact across all core security categories.

• Overall national security impact: Negligible
• Scope: Multiple nations
• Primary target: No clear primary
• Other affected: Austria, Japan, United Arab Emirates
• Alleged perpetrator: KPMG

Threat characteristics

• Imminence: Long-term. The incident represents an ongoing data quality and model pollution issue rather than an active security crisis requiring urgent response.
• Autonomy: Human-controlled. The AI tool was used as an assistant to generate references, while humans retained ultimate control and responsibility over the decision to publish.
• Novelty: Established threat. LLM hallucinations and the generation of plausible-sounding but fake citations are well-documented, established behaviors of generative AI models.

Impact by dimension

• Physical security: Negligible. No physical infrastructure, kinetic targeting, or human safety systems were affected or compromised by this corporate research incident.
• Information security: Negligible. The incident involves unintentional hallucinations in a corporate report rather than a state-sponsored disinformation campaign or a threat to intelligence capabilities.
• Sovereignty: Negligible. No government functions, electoral systems, or sovereign decision-making processes were impacted by this corporate reporting error.
• Economic security: Negligible. While it caused reputational risk for KPMG and misattributed companies, it does not threaten strategic industries, financial stability, or technological competitive advantage.
• Societal stability: Negligible. The incident does not involve mass surveillance, systematic discrimination, or any threats to civil liberties or societal stability.