Particle.news
Download on the App Store

ChatGPT Found Using xAI’s Grokipedia, Raising New AI Feedback-Loop Concerns

Experts warn that AI models can amplify errors when they learn from other AI-generated material.

Overview

  • The Guardian’s review of GPT-5.2 responses found answers that drew on Grokipedia for topics including Iranian politics and historical figures.
  • In several cases, the Grokipedia-origin material was not clearly credited as a source in ChatGPT’s citations.
  • Heise reports similar Grokipedia-derived passages appearing in Anthropic’s Claude, indicating cross-model seepage beyond OpenAI.
  • Grokipedia, launched by Elon Musk’s xAI and populated largely by the Grok model, has published disputed claims and demeaning language flagged by reporters.
  • OpenAI says ChatGPT relies on diverse sources with safety filters and programs to weed out low-credibility content, as experts point to data poisoning risks and influence campaigns targeting AI systems.