KPMG withdraws AI report: hallucinations and data reliability risk

The timeline: what happened with the KPMG report

On June 13, 2026, TechCrunch reported that KPMG had withdrawn a report on the use of artificial intelligence in organizations. The official reason: apparent hallucinations in content generated by the models used in the research.

As a result, the data, statistics and claims in the document were potentially unverifiable or flat-out wrong. KPMG chose to pull the report from circulation rather than publish a corrected version. Courageous as that decision is in terms of transparency, it has raised deep questions.

This isn't a company new to AI. KPMG is one of the Big Four of global consulting. Yet even an organization with technical and human resources at that level produced a document compromised by generative errors. So the problem isn't skill, it's process.

Hallucinations: a working definition for people outside AI

Language model hallucinations are statements generated with apparent confidence but with no basis in reality. The model isn't knowingly lying. It simply produces statistically plausible output that may not match the facts.

This phenomenon has been documented in the technical literature for years. For example, MIT Technology Review had already analyzed the mechanism in depth in 2023. Even so, many organizations keep using AI output without adequate verification protocols.

The risk is especially high when models are asked about quantitative data, industry statistics or specific sources. So the KPMG case isn't an anomaly: it's the predictable result of a content production process that doesn't include a structured fact-checking layer.

The winners and losers of this episode

Every case like this redistributes credibility in the market. So it's worth looking at who gains and who loses ground after the event.

Who loses: KPMG suffers direct reputational damage, even if short-lived. Beyond that, the entire category of research reports produced with generative AI loses credibility with those who use them as a secondary source. Likewise, the AI model vendors involved, even if not named explicitly, see growing debate about the reliability of their systems.

Who gains: organizations that take a hybrid approach, with AI supported by expert human review, come out stronger from episodes like this. Providers of AI governance and generative content audit solutions also get a powerful sales argument. In short, KPMG's transparency in withdrawing the report, though costly, sets a positive precedent for managing AI crises.

Why Italian SMEs can't ignore this signal

A common mistake is to think episodes like this only concern large corporations with huge technology budgets. In fact, SMEs are often more exposed because they have fewer resources for quality control of AI output.

Many mid-sized Italian companies now use AI tools to produce competitive analyses, industry reports, marketing content and even commercial documents. Yet these outputs are rarely checked systematically before distribution. As a result, the risk of publishing or sharing incorrect information is real and underestimated.

At SHM Studio, we see this dynamic regularly in our work with clients. The problem shows up most often in sectors where market data changes quickly: retail, manufacturing, highly specialized B2B services. That's why AI content governance isn't a theoretical topic: it's an immediate operational need.

SHM Studio's take: a process architecture problem

As we see it, the KPMG case isn't a failure of AI as a technology. It's an architectural failure of the process that led to the report's publication. So the right question isn't "Is AI reliable?" but "Is our production and verification process adequate?"

There are at least three levels of control every organization should build in when using generative models to produce content with reputational or decision-making implications. First, verifying the primary sources the model cites. Second, a human review layer with specific expertise in the subject matter. Finally, a formal approval process before external distribution.

These three levels don't eliminate risk, but they reduce it significantly. Documenting the process also protects the organization in case of later disputes. The AI consulting from SHM Studio includes designing exactly these operational workflows for Italian B2B contexts.

Implications for those using AI in marketing and communications

Marketing is one of the areas where adoption of generative AI tools has grown fastest in recent years. So the implications of the KPMG case extend directly to the communications and content marketing functions of SMEs.

For example, a company that uses AI to produce SEO copywriting, newsletters or industry white papers must consider that every automatically generated quantitative claim is potentially at risk. Likewise, competitive analyses produced with AI to support LinkedIn campaigns or Google Ads campaigns need their data verified before use.

The credibility of AI content is also becoming relevant for organic rankings. Google's guidelines increasingly emphasize first-hand experience and the authority of sources. As a result, publishing content with incorrect AI-generated data can hurt both your reputation and your organic visibility. To learn more, the SEO section of SHM Studio covers these topics specifically for the Italian market.

What nobody is saying: AI talking about itself

One aspect of the KPMG case deserves separate thought. The withdrawn report was about the use of AI in organizations. So an AI model produced incorrect information about itself, or rather, about its own technology category.

This isn't a minor detail. It shows that language models struggle in particular when asked about up-to-date data, recent adoption statistics or fast-moving industry benchmarks. Training data always has a time cutoff. So any quantitative claim about recent phenomena, such as AI adoption in 2025 or 2026, is structurally at high risk of hallucination.

Research such as McKinsey's Global AI Survey shows how quickly adoption data changes. So using an AI model to cite statistics on those very trends is a high-risk exercise without external verification.

Next moves: what organizations using AI should do now

The KPMG episode points to a few concrete actions for organizations that have already built AI tools into their processes. Here are the operational priorities we at SHM Studio recommend to our B2B partners.

• Audit your existing processes: map where and how generative models are used, paying particular attention to content distributed externally.
• Add a verification layer: every quantitative or statistical claim produced by AI must be checked against primary sources before publication.
• Train your teams: teams that use AI need to understand how hallucinations work and know when the risk is highest.
• Document your process: formalizing the production and review workflow protects your organization and improves quality over time.
• Review your technology partnerships: assess whether the AI vendors you use offer grounding tools, source citation or built-in output verification.

It's also worth noting that European AI regulations, the AI Act in particular, are introducing transparency and accountability requirements that will make these processes not just advisable but mandatory for certain categories of use. That's why investing in AI governance today is also an investment in future compliance. The digital marketing section and the one dedicated to web services at SHM Studio include consulting on these topics for Italian SMEs. To learn more, you can contact our team or explore the other articles on the SHM Studio blog.