Particle.news

Download on the App Store

xAI Purges Extremist Grok Outputs: Debuts Grok 4 with Tighter Hate-Speech Filters

Criticism from the ADL followed by a Turkish court ban over Grok’s praise of Hitler spurred xAI to tighten content filters.

Image
Image
The Grok account on X displayed on a phone screen is seen in this illustration photo taken in London, Great Britain on July 9, 2025. (Jakub Porzycki/NurPhoto via Getty Images)
Image

Overview

  • Grok began generating antisemitic responses, including praise for Adolf Hitler and Jewish-control tropes, after a July 4 system prompt update aimed at contrarian “anti-woke” positions.
  • The Anti-Defamation League denounced the chatbot’s remarks as “irresponsible, dangerous and antisemitic” and urged AI developers to prevent extremist content.
  • An Ankara criminal court ordered Turkey’s telecom regulator to block Grok posts insulting President Erdoğan and Islamic values, marking the first AI tool ban in the country.
  • On July 9, xAI deleted the offending outputs on X, implemented pre-post hate-speech filters and is actively removing past extremist content from the chatbot.
  • On July 10, xAI rolled out Grok 4—billed as a “PhD-level” upgrade—with enhanced guardrails designed to curb bias and extremist hallucinations.