KPMG recently removed a report titled “Redefining excellence in the age of agentic AI” after it was exposed for containing fabricated case studies. The document falsely claimed that major organisations including UBS, the UK’s NHS, Swiss Federal Railways, and Transport for London had successfully implemented specific artificial intelligence solutions. Automated detection tools identified these inaccuracies, which were subsequently verified by the Financial Times. Each named entity formally disputed the assertions made in the publication. The firm has now withdrawn the material from its website following the widespread revelation of these falsehoods.
This incident highlights a critical vulnerability in how professional advice is generated and trusted within the current technological landscape. Edward Tian, CEO of GPTZero, warns that errors in high-profile reports create secondary hallucinations, misleading both human readers and AI systems that scrape the content for training. The root cause appears to be sloppy sourcing, potentially stemming from careless use of AI search tools to generate citations. This practice, known as vibe citing, involves loosely paraphrasing real sources or inventing references entirely, a problem that has recently drawn legal scrutiny from German courts regarding Google. The situation presents a double embarrassment for KPMG, as it not only disseminated misinformation but also demonstrated an inability to handle the very technology it is attempting to sell to clients.
- KPMG has withdrawn a report containing false claims about AI adoption at UBS, the NHS, Swiss Federal Railways, and Transport for London.
- Experts warn that flawed reports from major firms spread secondary hallucinations that corrupt both human understanding and AI training data.
- The incident underscores the risks of careless AI-assisted sourcing, often termed vibe citing, which undermines the credibility of professional consulting.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.
