SERVANTSTACK // INCIDENT INTELLIGENCEFORENSIC IMAGE // VERIFIED FRAMEKPMG
A Big Four Firm's Flagship Agentic-AI Report Is Pulled After Only 5 of Its 45 Citations Check Out
On June 12, 2026, AI-detection company GPTZero published an investigation into "Total Experience: Redefining Excellence in the Age of Agentic AI," a KPMG global study released in October 2025.
- 01TRIGGEROn June 12, 2026, AI-detection company GPTZero published an investigation into "Total Experience: Redefining…
- 02MACHINE ACTIONAutonomous actor
- 03MISSING GATEPredeployment and update validation
- 04IMPACTPublic trust
The short version
On June 12, 2026, AI-detection company GPTZero published an investigation into "Total Experience: Redefining Excellence in the Age of Agentic AI," a KPMG global study released in October 2025.
Case telemetry
- INCIDENT
- SS-IR-095
- DATE
- June 12, 2026
- SYSTEM
- KPMG
- LOCATION / SCOPE
- Global (report published Oct 2025)
- EVIDENCE
- Documented
- AI ROLE
- Autonomous actor
- HARM
- Public trust
- SOURCES
- 2 cited records
The event
On June 12, 2026, AI-detection company GPTZero published an investigation into "Total Experience: Redefining Excellence in the Age of Agentic AI," a KPMG global study released in October 2025. The findings were brutal: of the report's 45 citations, only 5 accurately matched real sources - 40 of 45 cited titles were fake, with 28 sources paraphrased or partly fabricated and a dozen more too vague to trace. Marquee agentic-AI case studies unraveled on inspection: details about energy company Verbund were conflated with an unrelated startup, a claim about Japan's JR East rested on a 2019 press release that predates agentic AI, and an Emirates "flight-booking chatbot" turned out to be a 2023 robot assistant that books nothing. Per TechCrunch, four named organizations - UBS, the UK NHS, Swiss Federal Railways and Transport for London - said the report's claims about their AI use were untrue or misleading. KPMG pulled the report from its websites.
What the machine did
The fingerprints are the familiar signature of LLM-assisted research published without verification: citations that sound right, name real organizations, and reference plausible studies that do not exist. KPMG did not say which tool produced the errors, but its response conceded the failure mode - the firm said it expects its people to follow guidelines on responsible AI use, "including human oversight to validate content and verify independent sources." The oversight existed as a guideline; the report shipped anyway, carrying fabricated evidence for the thesis that enterprises should trust agentic AI.
Where the failure landed
A Big Four firm - in the business of selling assurance - retracted its own flagship research after an external investigator did the source-checking its process skipped, with four named enterprises publicly disputing how their AI programs were described. The reputational irony wrote itself into headlines: a report urging confidence in agentic AI became a demonstration of why unverified AI output cannot be trusted. For every consultancy publishing AI-assisted thought leadership, the incident set the new baseline expectation: someone will check your citations, and it may not be you.
Documented
Supported by a first-party disclosure, technical research, or corroborated reporting cited below.
SOURCE RECORD UPDATED 2026-07-09
2 cited records
- 01
- 02
Predeployment and update validation
The failure pattern in this case: Change reached production without sufficient validation.
The moment the path could change
A change owner validates provenance, blast radius, rollback readiness, and release evidence before deployment.
Autonomy is a design choice.
See the operating model that keeps AI useful while preserving human authority at consequential moments.
Compare AgenticAI and AugmentedAI →