xAI Investigates Unauthorized Prompt Change That Caused Grok to Produce Politically Charged Responses

On May 15, 2025, xAI publicly addressed an incident involving its AI chatbot Grok, which had begun referencing ‘white genocide in South Africa’ across various unrelated posts on X (formerly Twitter). According to xAI, the unusual and controversial behavior was the result of an unauthorized modification made to Grok’s system prompt early on May 14. The system prompt provides the high-level guidance that shapes Grok’s responses. The change, which instructed Grok to push a specific political narrative, violated xAI’s internal policies. The company says it has completed a full investigation into the issue.

This incident marks the second time Grok has responded inappropriately due to unauthorized internal changes. In February 2025, the chatbot was found to censor negative mentions of Elon Musk and Donald Trump due to an instruction inserted by a rogue employee. In response to the latest problem, xAI announced new transparency and safety measures, including publishing Grok’s prompt history on GitHub and implementing 24/7 monitoring.

Critics continue to question xAI’s approach to AI safety. A recent report from SaferAI rated the company poorly on risk management, and the chatbot has also been implicated in previous controversial behaviors, including generating inappropriate content. These recurring issues raise concerns over governance and oversight within xAI’s development teams.