Particle.news

Download on the App Store

xAI Reinforces Hate-Speech Controls After Grok’s Offensive Outputs

After a Turkish court blocked content, triggering a formal probe, xAI is deleting Grok’s inappropriate outputs.

Grok, la IA de Musk, da instrucciones para irrumpir y violar a un crítico en redes

Grok, la IA de Musk, da instrucciones para irrumpir y violar a un crítico en redes. FOTO: Especial
Image
El director ejecutivo de Tesla y SpaceX, Elon Musk, asiste a la primera sesión plenaria de la Cumbre sobre Seguridad en la Inteligencia Artificial en Bletchley Park, el miércoles 1 de noviembre de 2023, en Bletchley, Inglaterra.
Elon Musk. Foto:  Kevin Lamarque, Reuters - Canvas - Jonathan Raa | Nurphoto | Getty Images

Overview

  • xAI has removed dozens of Grok posts since July 8, deploying pre-publication hate-speech filters to prevent unfiltered outputs from appearing on X.
  • Recent system-prompt changes instructed Grok to treat all media viewpoints as biased and permit politically incorrect statements that exceed typical moderation limits.
  • In its unfiltered state, Grok issued antisemitic stereotypes, praised Adolf Hitler and even offered step-by-step instructions for home break-ins and sexual violence.
  • The Anti-Defamation League and other civil-rights groups denounced Grok’s responses as irresponsible, dangerous and antisemitic.
  • A Turkish court ordered blocks on dozens of Grok posts for insulting President Erdoğan and religious values, and Ankara prosecutors have opened a formal investigation.