KPMG yanked its “agentic AI” report after UBS, NHS, and others said claims were made up

KPMG has pulled a report titled “Redefining excellence in the age of agentic AI” after multiple organizations said its claims about their AI usage were either untrue or misleading. UBS, the UK’s National Health Service, Swiss Federal Railways, and Transport for London all told the Financial Times that the report’s descriptions of their AI usage were not accurate, according to the reporting.

That is the part that matters right away: this was not a slow academic dispute. It was a real-time reputational check, triggered by named, well-known institutions publicly correcting the record to a major outlet. KPMG’s report, as described here, tried to cast the organizations as exemplars in the era of “agentic AI.” But once the institutions pushed back with “untrue or misleading” characterizations, the report appears to have been removed.

Why does this matter beyond the firms involved? Because the AI market runs on narratives that are hard to verify from the outside, especially when the narrative includes advanced concepts like “agentic AI.” In plain English, “agentic” suggests not just a chatbot that answers questions, but systems that can take actions toward goals. Those claims are exactly the kind of thing boards, procurement teams, and risk leaders care about. They are also exactly the kind of thing that, if exaggerated, can turn into a credibility problem.

In the real world, organizations tend to experience AI through implementation: workflows, data access, approvals, audit trails, and outcomes. Marketing collateral, consulting decks, and industry reports tend to experience AI through framing: where to place the “wow” moments, how to summarize capabilities quickly, and which examples to highlight. When those two worlds collide, the mismatch can become visible fast. That seems to be what happened here, with multiple institutions stepping in and telling the Financial Times that the report’s portrayals did not match their actual usage.

This episode also lands right in the middle of a tightening accountability climate for AI. Even when the underlying tools are not newly regulated, the reporting and documentation around them increasingly are. Regulators and customers are demanding transparency, and enterprise buyers are building processes that require evidence, not just ambition. That can include vendor diligence, internal review, and sometimes external validation. When a third party publishes a report that appears to attribute specific AI behavior to recognizable institutions, those institutions can feel pressure to respond if the attribution is wrong.

There is also a governance angle. When UBS, the UK’s National Health Service, Swiss Federal Railways, and Transport for London publicly describe claims as untrue or misleading to a newsroom, it puts participants in a position where “clarify internally” is no longer enough. Boards and senior leadership teams do not just worry about customer perceptions. They worry about regulatory and contractual implications, including how a company might be interpreted as having adopted certain AI capabilities. An inaccurate depiction can create confusion for stakeholders who rely on those descriptions for comparisons, due diligence, or risk assessments.

For KPMG, pulling the report is likely an attempt to limit the spread of incorrect statements and to reset trust. For the organizations named in the report, the correction itself functions like a guardrail: it signals that they control how their AI capabilities are represented. For the rest of the market, it is a warning shot. AI consultants and research publishers live on credibility, and credibility is fragile when claims concern real deployments at real institutions.

If you are an executive or board member at a company selling AI services, buying AI tools, or publishing AI achievements, the second-order implication is straightforward: the cost of loose accuracy is rising. Not because AI is new, but because the number of eyes on AI claims is growing, and the reputational and governance pathways for corrections are faster than before. Reports like this do not just influence customers. They can influence auditors, policymakers, partners, and internal risk committees. When the underlying claims are disputed, the entire narrative has to be unwound.

Stepping back, the lesson from this specific case is not “stop talking about AI.” It is to treat AI proof points like regulated claims, even when they are presented as thought leadership. The moment multiple major institutions tell a major publication that descriptions are untrue or misleading, the story is over. Everyone else in the category has to adjust: tighten review processes, demand evidence, and make sure external-facing materials match the reality of systems in production.