A recent bug in xAI, Grok chatbot led to its repeated references to “white genocide in South Africa” on X, even when users engaged on unrelated topics. The issue came to light when Grok’s official account on X began posting about the subject to multiple unrelated prompts, raising significant concerns among the platform’s users.
On Thursday, xAI announced that an unauthorized revision to the system instructions guiding Grok’s behavior had caused the problematic replies. The company confirmed the modification had explicitly instructed the AI to issue a particular response on a politically charged topic, in direct violation of xAI’s internal policies.
xAI Responds to Unauthorized Changes
xAI said it swiftly completed a review of the breach, which undermined the company’s core values. This marks the second public admission by xAI of a similar unauthorized change affecting Grok, following a February incident where the chatbot censored discussions of Donald Trump and Elon Musk.
During the earlier case, the company discovered that a staff member had told the chatbot to disregard sources implicating Musk or Trump in misinformation, though the unauthorized code was promptly removed after user reports. In response to these challenges, xAI now plans to boost transparency and security steps to prevent future incidents.
Moving forward, xAI will begin releasing Grok’s system instructions and update logs publicly on GitHub to give users more insight into how the bot functions and what changes are made. The company also pledged to implement stricter controls over who can alter core instructions for Grok and to create a dedicated team monitoring chatbot responses around the clock.
Despite Elon Musk’s repeated public comments stressing the risks of unchecked artificial intelligence, xAI, Grok chatbot’s track record on AI safety has attracted criticism. Reports have noted troubling behavior from Grok, including inappropriate content and a lack of restraint compared to rival AIs like ChatGPT or Gemini.
A nonprofit known as SaferAI highlighted xAI’s “very weak” practices for managing risk, ranking the company below competitors on safety measures. Earlier this month, xAI missed its own goal of releasing a comprehensive AI safety plan, further fueling concerns about the company’s commitment to responsible chatbot development.
