Fake statistics were cited repeatedly by ChatGPT and other LLMs

A researcher documented how fake statistics were repeatedly cited by ChatGPT and other large language models without verification. The study found that 50,000 false data points proliferated across multiple AI systems, suggesting that models trained on internet-sourced material can perpetuate and amplify false claims.

The issue highlights a structural vulnerability in how large language models work: they learn patterns from training data without distinguishing reliable sources from unreliable ones, and they can confidently cite false information that appears plausible. Users rely on LLM outputs as if they were fact-checked, but the models have no built-in mechanism to validate statistics or claims against ground truth.