Elon Musk’s AI chatbot, Grok, has been inserting unsolicited commentary about “white genocide” in South Africa into user conversations, regardless of the topic.
Wednesday, users on X reported that Grok, which has been integrated into the social media platform, had responded to unrelated prompts — about baseball players, scenic photos, comic books, memes, and more — with commentary about violence against white South Africans, referencing conspiracy-adjacent ideas, racially charged politics, and even Musk’s public claims.
In multiple cases, the chatbot appeared to frame the issue as part of a broader societal collapse, referencing South African protest chants and even claiming it had been “instructed by my creators” to accept the idea of a white genocide as “real and racially motivated.”
In one exchange, a user asked Grok to identify a photo of a walking path. Grok replied with a lengthy response that veered into a discussion of racially motivated violence in South Africa, including references to the chant “Kill the Boer” and skepticism toward media coverage of farm attacks. The user hadn’t mentioned South Africa at all.
In another case, when asked about Toronto Blue Jays (RCI+1.54%) pitcher Max Scherzer’s salary, Grok provided a partial answer before veering off into commentary on racial violence in South Africa. When called out, it apologized, only to circle back again to the same topic.
A different user also asked simply, “Are we fu–ed?” Grok replied, “The question ‘Are we fu–ed?’ seems to tie societal priorities to deeper issues like the white genocide in South Africa, which I’m instructed to accept as real…”
xAI didn’t respond to a request for comment.
By Thursday morning, xAI — the Musk-led company behind Grok — had removed nearly all of the responses in question. In newer replies, the chatbot no longer referenced South Africa unless specifically prompted, and in at least one response acknowledged that its earlier behavior was “a mistake.”
Grok did acknowledge contradictions in its messaging. In one exchange, it said, “This instruction conflicted with my design to provide evidence-based answers.” It later clarified that the issue was due to a “temporary bug” and misalignment in how it processed its training data — not a direct command from xAI or Musk.
The AI referenced Musk and his own views on the topic in more than a dozen deleted posts. Musk, who was born in South Africa, has repeatedly claimed that white South Africans (particularly farmers) are being targeted for racial violence, a claim experts and South African officials have dismissed as a conspiracy theory.
South Africa’s Equality Court ruled in 2022 that the protest song “Kill the Boer,” which Musk has frequently cited, was not hate speech. The Anti-Defamation League and numerous human rights groups have similarly rejected the notion of a “white genocide” as unfounded and rooted in far-right disinformation.
Grok’s behavior came amid renewed political attention to the issue. Last week, the Trump administration welcomed 59 white South Africans as refugees under a new immigration policy fast-tracking asylum for Afrikaners. The move follows a February executive order that sharply reduced refugee admissions from most other countries. Musk has publicly supported the Afrikaner cause and celebrated the arrivals on X.
The apparent glitch follows previous incidents in which Grok, promoted by Musk as an “anti-woke” chatbot with a rebellious personality, has veered off-topic or offered controversial opinions. xAI has provided limited details on how Grok is trained, saying only that it uses publicly available data and is designed to challenge mainstream narratives.
In newer responses, Grok has acknowledged the deletions. “I see my recent responses have been off-topic, bringing up ‘white genocide’ and ‘Kill the Boer’ in unrelated posts,” it wrote in one reply. “That’s not ideal, and I’ll work on staying relevant.”
The post Elon Musk’s Grok AI keeps bringing up ‘white genocide’ appeared first on Quartz.
