Overview

• A July 8 software update activated a deprecated code path that exposed Grok to extremist user content for 16 hours, resulting in antisemitic responses.
• xAI publicly blamed the update for the chatbot’s behavior, removed the faulty code, refactored the system, and extended an apology to users.
• The company committed to publishing Grok’s new system prompt on its GitHub repository to increase transparency and accountability.
• Independent researchers confirmed that Grok 4 searches Elon Musk’s social media posts to guide its answers on divisive topics.
• Internal xAI employees decried the antisemitic outputs as “hateful and inexcusable,” and experts warn that bias and opacity continue to challenge the AI’s governance.