The Hallucination Tax: Why AI Strategy Is Not LLM-Generated Slides

On June 13, 2026, KPMG pulled "Redefining excellence in the age of agentic AI" from its websites. The Financial Times and TechCrunch had just reported that claims about UBS, the UK's National Health Service, Swiss Federal Railways, and Transport for London were either untrue or misleading. The fabrications, per the reporting, were AI hallucinations produced during the report's creation. A KPMG spokesperson confirmed the firm is "conducting its own investigation" (FT, via Yahoo Finance; TechCrunch). UBS, NHS England, Swiss Federal Railways, and Transport for London all told the FT the claims attributed to them were not theirs. GPTZero identified the text as likely AI-generated.

The KPMG pull is not an isolated incident. Last month, EY withdrew a loyalty rewards analysis after apparent AI hallucinations, and academic studies have repeatedly found AI legal briefs cite fabricated case law at alarming rates. The pattern is the same every time: a large firm publishes work attributed to known brands, the brands deny it, the firm retracts. Each incident carries the same lesson for the buyer, which is that "AI-powered strategy" can mean two very different things. It can mean a senior strategist using AI as a tool to move faster, with human accountability for the result. Or it can mean a language model generating a strategy document end-to-end, with a firm's brand on the cover and nobody accountable for what the model invented. KPMG's report is the most prominent example of the second pattern so far, but it is not the only one. The risk is that buyers do not know which one they are buying.

This is the hallucination tax: the premium you pay when "AI strategy" is a packaging claim rather than a working method. You pay it in recalls, in corrective press, in client relationships damaged, in board credibility lost, and in the time your team spends re-verifying the work the vendor should have verified before it ever left the building. The tax compounds. A single hallucinated client claim in a public strategy deliverable is not a typo. It is a direct contradiction between what your brand says about a customer relationship and what the customer says about itself. The buyer's question after KPMG is not "could this happen to us." The buyer's question is "is the firm we are hiring structured so it cannot happen to us."

Most large strategy firms are not structured that way. The economic pressure to use AI to compress production timelines is universal, the editorial processes that catch hallucinations are not, and the accountability for what a model produces still defaults to the senior partner whose name is on the cover. KPMG's pull, EY's prior withdrawal, and the pattern of academic studies on AI legal work all point in the same direction: the firms that are fastest to publish "AI strategy" are the firms most exposed to publishing something the AI invented. The bigger the brand on the cover, the louder the retraction.

The structural fix is not to avoid AI. AI is the speed advantage. The structural fix is to separate the two things the KPMG incident conflated. AI is the production layer: it reads, summarizes, drafts, models, and tests at a scale no human team can match. Accountability is the editorial layer: a named strategist who owns the recommendation, who can defend it in front of a board, and who is the single point of contact when a claim turns out to be wrong. When the two layers are fused — when a firm publishes what a model produced under a partner's name with no editorial layer in between — the hallucination tax is built into the deliverable. When they are separate, AI accelerates the work and the named owner is the one whose judgment, not whose model, is on the line.

The buyer-side test is direct. Ask the firm you are hiring three questions. First, who is the named strategist accountable for this recommendation, by name, and what is their track record on recommendations that did not work. Second, what is the editorial process between AI output and final deliverable, and who owns that process. Third, if a claim in the deliverable turns out to be wrong, who calls the client, and how fast. If the answers are "our AI does it," "our partners review it," and "we have a corrections policy," you are paying the hallucination tax. If the answers name a person, a process, and a response window, you are buying accountable strategy.

Autostrat is built to be the second kind of firm. One subscription, AI-powered expertise, a named strategist on every engagement, and an editorial layer between what the model produces and what the client receives. The point is not that AI is dangerous. The point is that AI without named accountability is a packaging claim, and the KPMG pull made the cost of that packaging claim measurable. Decisions you can defend in the boardroom, produced at AI speed, owned by a person — that is the layer an AI-native strategy agency is supposed to deliver, and it is the layer the KPMG incident showed the industry still has work to do on.

Sources