Recently, the chatbot Grok began answering any questions X users had about the “white genocide” in South Africa. These answers appeared even for the most off-topic questions. Elon Musk’s startup xAI, which is developing the chatbot, said that this behavior was caused by “unauthorized modification.”
Similar responses emerged when X users used Grok to clarify or verify information written in other posts. In one case, the chatbot started talking about white genocide when asked how many times HBO had changed the name of its streaming service, which had happened again the day before. In another case, Grok talked about white genocide when asked to explain a post by the new Pope in Fortnite terminology, or to check the salary of an American baseball player.
xAI has now removed all such responses from the chatbot and issued a statement saying that unauthorized changes were made to the bot that required it to provide a specific response on a political topic. The startup notes that a thorough investigation was conducted and necessary measures were taken.
In particular, xAI will start publishing Grok system hints in the public domain on GitHub, where users can review them and provide feedback on every operational change made to the chatbot. In addition, a monitoring team has been created that will work 24/7 and respond to such incidents. The startup also notes that in this case, the existing code review process for operational changes was bypassed, so the developers will implement additional checks and measures to ensure "that xAI employees cannot change the hints without review."
At the same time, xAI does not say who exactly made the "unauthorized modifications" that allowed the chatbot to spread information about the far-right conspiracy theory of "white genocide" in South Africa, and how. Elon Musk, who is originally from South Africa, has previously spread similar messages on his social networks. They have also been spread by figures such as Tucker Carlson, and Donald Trump previously granted asylum to 59 white South Africans, speeding up the process of obtaining refugee status for them, while thousands of refugees from other countries wait for years.
It is worth noting that this is not the first time that Grok has spread information that plays into the hands of Elon Musk. In February 2025, the xAI chatbot temporarily refused to answer questions that mentioned that sources associated with Musk and Donald Trump were spreading misinformation. At that time, this was also explained by unauthorized changes to Grok's hint system made by one of the company's employees.