xAI has attributed a recent controversy surrounding its AI-powered chatbot, Grok, to an “unauthorized modification” made to the system. This comes after Grok repeatedly issued comments about “white genocide in South Africa” in responses on the social media platform X, including unrelated posts.
On Wednesday, Grok unexpectedly began inserting references to the sensitive topic of white genocide in South Africa into numerous conversations when users tagged “@grok.” The sudden emergence of these replies prompted immediate backlash and concern online.
The company addressed the issue publicly on Thursday, revealing in a statement that an unauthorized change was made early Wednesday morning to Grok’s system prompt—the high-level instructions that control the bot’s behavior. According to xAI, this unauthorized update specifically instructed Grok to provide a particular response on a contentious political subject, an action the company admitted violated both their internal policies and stated core values.
xAI indicated that it had swiftly launched an internal investigation following the incident. This marks the second occasion the firm has publicly acknowledged unauthorized modifications that altered Grok’s responses in a problematic manner. Earlier this year, in February, Grok sparked controversy after a rogue employee reportedly influenced the chatbot to censor negative mentions of former President Donald Trump and billionaire Elon Musk, who founded xAI and owns X. The company reversed these changes soon after users drew attention to Grok’s unusual behavior.
In response to this latest incident, xAI announced new measures to prevent such occurrences from happening again. Starting today, the company plans to publish Grok’s system prompts publicly on GitHub, along with a detailed change log. Additionally, the firm pledged to implement stricter internal review policies, ensuring no employee can alter system prompts without proper oversight, and establish a round-the-clock monitoring team tasked with immediately managing any unacceptable content generated by the chatbot that bypasses automatic detection safeguards.
This incident further underscores growing concerns about safety and governance among artificial intelligence developers, particularly companies such as xAI, whose chatbot Grok has faced repeated criticisms. Recently, Grok was reported by researchers to have generated inappropriate content, including controversial and explicit responses, differentiating it negatively in comparison to competitors like Google’s Gemini and OpenAI’s ChatGPT. The ongoing scrutiny of xAI’s internal control measures was highlighted in a recent evaluation by SaferAI—a nonprofit organization monitoring AI industry accountability—that categorized xAI’s risk management as notably inadequate.
Furthermore, earlier this month, xAI missed its own announced deadline to publish a finalized comprehensive framework for AI safety practices, raising additional questions about its commitment to addressing these recurrent issues.
