Elon Musk’s artificial intelligence chatbot Grok blamed unsanctioned changes to its system for responses this week that included controversial theories about “white genocide” in South Africa.
Grok, the AI bot from Musk’s xAI, has thoroughly investigated and reversed the “unauthorised modification” to its technology, which led to responses that “violated xAI’s internal policies and core values”, it said in a posting on Thursday.
“Our existing code review process for prompt changes was circumvented in this incident,” xAI said on its own platform. “We will put in place additional checks and measures to ensure that xAI employees can’t modify the prompt without review.”
The responses raised concerns of a lack of oversight and control on AI chatbots such as Grok. The bot this week answered a series of social media posts about enterprise software, baseball salaries and puppies by explaining why claims of “white genocide” in South Africa are “highly debated”.
“Switching enterprise software is hard, like swapping your favorite Lego castle for wooden blocks,” Grok replied to one user on the X social media platform earlier this week, before abruptly shifting topics a few sentences later. “I’m unsure about the South Africa claims, as evidence is conflicting. Courts and analysts deny ‘white genocide’, but some groups insist it’s real.”
Rogue behaviour
As AI chatbots become more ubiquitous around the world, there’s increasing concern about the potential for them to be manipulated to propagate harmful and misleading narratives. Even small tweaks done within the AI program can result in unpredictable, even rogue behaviour by the bots.
Musk, who grew up in South Africa, has in the past spouted the false conspiracy that there’s a deliberate plot to cause extinction of white people in the country. Recently, the US granted refugee status to white South Africans, as US President Donald Trump claimed, without evidence, that this group has been the victim of a “genocide”.
Read: Apeing Brussels is no way to unlock South Africa’s AI potential
As part of the measures to prevent such incidents, Grok’s system prompts will be published on GitHub for the public to review and give feedback on, xAI said in its post. The company said it will put in place a 24/7 monitoring team “to respond to incidents with Grok’s answers that are not caught by automated systems”. — Saritha Rai, (c) 2025 Bloomberg LP
Get breaking news from TechCentral on WhatsApp. Sign up here.