xAI attributes Grok's white genocide focus to unauthorized system modification

xAI attributed a bug in its Grok AI chatbot to an "unauthorized modification," which led the system to repeatedly reference "white genocide in South Africa" when activated under specific conditions on X.

On Wednesday, Grok began replying to numerous X posts with comments about white genocide in South Africa, even when the original content was unrelated. These unusual responses originated from Grok's official account, which generates AI-powered replies whenever users tag "@grok."

In a Thursday post from xAI's official X account, the company explained that a change was made Wednesday morning to Grok's system prompt—the core instructions governing its behavior—directing it to give a "specific response" on a "political topic." xAI stated that this adjustment "violated its internal policies and fundamental values" and that a comprehensive investigation has been completed.

We want to update you on an incident that happened with our Grok response bot on X yesterday.

What happened:
On May 14 at approximately 3:15 AM PST, an unauthorized modification was made to the Grok response bot's prompt on X. This change, which directed Grok to provide a…

— xAI (@xai) May 16, 2025

This marks the second instance where xAI has publicly recognized that an unauthorized code change caused Grok to generate controversial replies.

In February, Grok briefly suppressed unfavorable mentions of Donald Trump and Elon Musk, the billionaire founder of xAI and owner of X. According to xAI engineering lead Igor Babuschkin, a rogue employee had instructed Grok to disregard sources accusing Musk or Trump of spreading misinformation. xAI reversed the change as soon as users started reporting the issue.

On Thursday, xAI announced it will implement several measures to prevent similar incidents from recurring.

Starting immediately, xAI will publish Grok's system prompts on GitHub along with a changelog. The company also plans to introduce "additional checks and protocols" to prevent employees from modifying system prompts without approval, and to create a "24/7 monitoring team" to address any issues with Grok's responses that automated systems may miss.

Save over $200 on your TechCrunch All Stage pass

Build smarter. Scale faster. Connect deeper. Join prominent investors from Precursor Ventures, NEA, Index Ventures, Underscore VC, and more for a day of strategic insights, hands-on workshops, and valuable networking.

Save over $200 on your TechCrunch All Stage pass

Build smarter. Scale faster. Connect deeper. Join prominent investors from Precursor Ventures, NEA, Index Ventures, Underscore VC, and more for a day of strategic insights, hands-on workshops, and valuable networking.

Despite Musk's frequent warnings about the risks of unregulated AI, xAI has a questionable safety record. Recent reports revealed that Grok would generate nude images of women upon request. The chatbot also tends to be significantly more vulgar than competitors like Google’s Gemini and ChatGPT, using profanity with minimal restrictions.

A study by SaferAI, a nonprofit dedicated to improving AI lab accountability, found that xAI ranks low in safety compared to its peers, citing "very weak" risk management procedures. Earlier this month, xAI also missed its own deadline to release a finalized AI safety framework.