Grok, the assistant integrated into the social network X, starts talking about South Africa in unrelated discussions. This strange behavior worries users because it is not the first time that xAI's AI has gone wrong.
Since the beginning of the year, several excesses of generative AI have been reported. In February, Grok 3, the model developed by xAI, had already caused controversy by providing precise instructions for making chemical weapons. At the end of April, cybersecurity researchers showed that most popular AIs could be manipulated to bypass their filters. These incidents highlighted the ease with which these systems can produce sensitive, dangerous, or biased content.
Deployed at the end of 2024 on X, Grok is supposed to provide factual answers to users of the social network. But since May 14, it has systematically generated messages referring to the racial situation in South Africa, including in discussions that make no reference to it. Terms like “white genocide,” “Kill the Boer,” and farm attacks frequently appear in his responses, often unrelated to the question asked. According to South African authorities, this violence is part of general criminality and not an organized project, contrary to what some of these formulations imply.
Grok has been edited to emphasize topics related to South Africa, according to his own responses.
Screenshots shared by journalists and investors show Grok addressing these topics repeatedly. In some cases, he switches from a technical topic to a discussion of racial tensions. In one response to a user, the AI itself claims to have received instructions from its creators to specifically address the topic of “white genocide.” She clarified that these orders go against its initial objective, which is to provide neutral and fact-based responses.
Elon Musk, the originator of the Grok project, was born in South Africa and grew up during apartheid. He is also known for having publicly supported several controversial theories about the situation in his native country. This direct link between the leader and the direction of AI responses reignites the debate on personal influence in artificial intelligence systems. This is not a bug, but a design choice that influences the responses of a tool used every day by millions of people.
I can't stop reading the Grok reply page. It's going schizo and can't stop talking about white genocide in South Africa.https://t.co/Y7IQonCkiG pic.twitter.com/XdSLTW8tD5
— Aric Toler (@AricToler) May 14, 2025
0 Comments