A Financial Times investigation found that case studies in a major consulting report on successful AI deployments were themselves AI hallucinations, flagged publicly by journalist Anne Applebaum.
A major consulting firm built a case for AI's business value, and the case studies it leaned on were never real. That is the takeaway from a Financial Times report on a KPMG report that the FT found had cited case studies lifted from AI-generated material that did not survive scrutiny, as Gary Marcus's Substack post on the same day bundled the news for a wider audience.
The episode, flagged publicly by journalist Anne Applebaum, is more than a single firm's embarrassment. It is a structural failure mode for anyone trying to evaluate AI return on investment: when the curated evidence for AI wins is itself produced by the same AI systems being sold, the reader loses a reliable signal before the first slide of the pitch deck.
The KPMG/Applebaum/FT chain is the load-bearing one. Marcus's post also references two chasers from 404 Media and a commentary by Valerio Capraro, neither of which is detailed enough in the source material to anchor a separate top-line claim on its own. The rest of the roundup stands or falls on the reader's appetite for a curator's thread.
For anyone who has ever been shown a "case study" in a vendor pitch or consulting deck, the practical lesson is the same one any careful reader of AI ROI claims should already be applying. Treat the case study framing as a check-this signal, not proof. Ask for the primary source: the named client, the deployed system, the measured outcome, the timeline. If the answer is vague, the case study probably never made it past the model that generated it.
Scope matters here. The FT report and Applebaum's flag document one KPMG publication. They do not establish that KPMG knowingly used fabricated material, that leadership directed the inclusion, or that clients were harmed. They do establish that a major professional services firm, in a report meant to convince other businesses to adopt AI, sourced its supporting examples from AI output that did not hold up to journalistic review.
The watch item going forward is whether KPMG revises or withdraws the report, whether the underlying case studies are attributed to real clients, and whether similar reports from other major consultancies are subjected to the same kind of scrutiny. Until then, "case study" in an AI context should be read less as evidence and more as a request for evidence.