In mid-June 2026, KPMG withdrew one of its reports on the use of artificial intelligence. The reason: the document contained data that was apparently generated by AI systems with evident hallucinations. Therefore, the published information was neither verifiable nor reliable.
This episode is not isolated. In fact, it represents a systemic signal for all organizations using AI tools in enterprise analytics. However, the problem doesn't only concern large consulting firms. Italian SMEs that rely on unsupervised AI outputs also risk building strategic decisions on incorrect data. Consequently, the governance of AI-generated content becomes an operational priority, not a theoretical issue.
We of SHM Studio We are closely following these developments. In particular, we are working with Italian SMEs to integrate AI into marketing and communication processes in a controlled and verifiable manner. Finally, this KPMG case offers a concrete insight for reviewing internal workflows and strengthening fact-checking processes for every output generated by artificial intelligence.
KPMG Case History
On June 13, 2026, TechCrunch reported KPMG has withdrawn a report dedicated to the use of AI. The document, intended for an enterprise audience, contained unverifiable data and claims. The identified cause: hallucinations produced by the artificial intelligence systems used in its drafting.
The withdrawal happened quietly, but the news spread rapidly within the industry. Indeed, KPMG is one of the Big Four global consulting firms. Therefore, an error of this magnitude carries significant symbolic weight. This isn't a startup experimenting; it's an organization with structured internal processes.
However, the situation is not surprising to those who closely follow the evolution of language models. Generative AI systems tend to produce plausible output even in the absence of real data. This phenomenon, known as hallucination, has been documented for years in technical literature.
Winners and losers in this matter
Who is harmed by this episode? First and foremost, KPMG itself. A consulting firm's reputation is built on the quality and accuracy of its analyses. A report withdrawn due to AI hallucinations weakens the trust of institutional clients.
Secondly, the entire ecosystem of enterprise AI vendors suffers. Many of these tools are sold as reliable solutions for report generation and analysis. Consequently, episodes like this fuel skepticism in decision-makers. Among other things, some organizations might slow down their adoption plans.
Who, on the other hand, can benefit from this situation? Companies and agencies that have built human oversight processes for AI output. Furthermore, professionals who promote a hybrid approach—AI as a tool, human as a validator—see their position strengthened. Likewise, AI governance frameworks gain concrete relevance today that until yesterday was more theoretical.
Why do hallucinations affect analytical reports specifically?
Large language models do not reason; they generate statistically coherent text based on the prompt received. Therefore, when questioned about specific data—percentages, statistics, research—they tend to produce plausible numbers even if there are no real sources to support them.
This mechanism is particularly insidious in analytical reports. In fact, in a document of this type, an invented but correctly formatted piece of data can pass superficial reviews. In particular, if the reviewer is not familiar with the original source, verification becomes very difficult.
According to research from Gartner on Large Language Models, The management of hallucinations remains one of the main challenges for the enterprise adoption of generative AI. Therefore, it is not a residual problem destined to disappear with future model updates.
In addition to this, the problem is amplified when AI is used to analyze data about AI itself. In this case, the models draw on a training corpus that may contain contradictory, obsolete, or AI-generated information. The result is a potentially self-referential error loop.
SHM Studio Reading: A Governance Problem, Not a Technology Problem
We of SHM Studio We interpret the KPMG case as a process problem, not a tool problem. Generative AI is not inherently unreliable. However, it becomes dangerous when inserted into workflows lacking verification checkpoints.
The correct question is not
Related articles
Discover other articles that explore similar topics in depth, selected to give you a more complete and stimulating view. Each piece of content is carefully chosen to enrich your experience.
