𝗞𝗣𝗠𝗚 𝗨𝘀𝗲𝗱 𝗔𝗜 𝗧𝗼 𝗪𝗿𝗶𝘁𝗲 𝗔 𝗥𝗲𝗽𝗼𝗿𝘁 𝗔𝗯𝗼𝘂𝘁 𝗔𝗜. 𝗜𝘁 𝗙𝗮𝗶𝗹𝗲𝗱.
KPMG sells advice on how to use AI responsibly.
In October 2025, they published a report on Agentic AI. They used AI to write it.
The results were a disaster.
An audit by GPTZero found that the AI invented 88% of the sources.
Here are the facts from the KPMG report:
- Total citations: 45
- Real citations: 5
- Invented citations: 40
- Accuracy rate: 11.1%
The report claimed major companies like UBS and the NHS were using Agentic AI. These companies said the claims were false or misleading. KPMG used their logos on fiction without permission.
This is a new problem called "vibe citing."
Vibe citing happens when you use AI to generate a bibliography. The AI creates references that look perfect. The titles sound academic. The dates look right. The format is correct. But the content is fake.
The AI does not know the difference between remembering and inventing. It only knows how to generate what fits the pattern.
This is not just a technical error. It is a process failure.
KPMG has 250,000 employees. They published a report with 11% accuracy. They failed to verify a single source before hitting publish.
The danger goes beyond one bad report.
Because KPMG is a high-authority source, web crawlers indexed this fake data. Now, other AI models like ChatGPT and Gemini ingest this false data. They treat it as truth.
The cycle looks like this:
- AI hallucinates data
- A major company publishes it without checking
- Search engines index it as a fact
- Other AI models learn it as truth
- Users receive hallucinations as facts
AI is not the problem. The problem is the pressure to appear knowledgeable without doing the work.
Using AI without a human verification loop is not efficiency. It is delegating truth to a system that has no concept of truth.
Stop appearing to know. Start knowing.
Source: https://gptzero.me/news/investigations-kpmg/
Optional learning community: https://t.me/GyaanSetuAi