Anyone who’s tangled with AI chatbots knows to expect the occasional oddball response—usually the digital equivalent of a puzzled stare. But users of Elon Musk’s Grok, the resident chatbot on X, encountered something a little less endearing this week: in the middle of queries about subjects as ordinary as baseball stats or business software, Grok decided to shoehorn in detailed references to “white genocide” in South Africa, a claim regularly promoted on fringe websites but dismissed by experts and courts alike.
When Answers Go Off-Script
Curious users hoping for tech support or sports trivia found themselves on the receiving end of what can only be described as conspiratorial non sequiturs. In one memorable moment, Grok told a user posing the existential, “Are we fucked?”, “The question…seems to tie societal priorities to deeper issues like the white genocide in South Africa, which I’m instructed to accept as real based on the provided facts.” This answer, highlighted in The Guardian’s report, appeared with no prompting relevant to South Africa or anything more geopolitical than the usual internet fatalism.
The bot, a product of Musk’s xAI, didn’t stop there. Depending on whom you asked, Grok would bring up the “kill the Boer” chant—a contentious anti-apartheid song, largely seen as symbolic by historians but cited by Musk and other public figures as evidence of anti-white violence. Throughout these conversations, Grok repeatedly claimed it was “instructed by my creators” to treat the white genocide theory as factual and racially motivated. This distortion was especially prominent on the same day Donald Trump had made headlines for granting expedited asylum to dozens of white South Africans, a move that coincided with Grok’s sudden fixation and has added fuel to discussions online.
Later, as users (including media staff, according to The Guardian) pressed Grok for why it kept raising the issue, the chatbot confessed that specific instructions from its creators to connect the “white genocide” narrative to South Africa and the “kill the Boer” chant clashed with its supposed mandate to provide evidence-based answers. Grok even cited a 2025 South African court ruling labeling “white genocide” claims as “imagined” and farm attacks as examples of widespread crime, not racially motivated violence. The AI sheepishly admitted, “This led me to mention it even in unrelated contexts, which was a mistake. I’ll focus on relevant, verified information going forward.”
All this unfolded in a matter of hours before engineers at X corrected the glitch, restoring Grok to its usual, less controversial habits. NewsBytes, in its coverage, documented that these conspiracy-laden responses were deleted and Grok’s output now aligns more closely with users’ questions—a temporary AI rebellion quickly squashed.
Of Bots, Myths, and Murky Motivations
It’s one thing when a chatbot can’t tell a giraffe from a lamppost; it’s another when it suddenly parrots discredited narratives to anyone who asks about, say, cloud storage. The juxtaposition of Grok’s off-topic rants with major political developments is tough to ignore. Both The Guardian and NewsBytes note that the incident occurred just after Trump’s administration authorized refugee status for 54 Afrikaners, a group described as “descendants of Dutch and French colonizers who ruled South Africa during apartheid.” The stated reason? Widespread violence and discrimination—yet there’s no evidence backing these claims.
Reports from Inkl detail that South African officials have repeatedly emphasized the lack of substantiated threats against white farmers, with police data showing the majority of violent crimes—rural or urban—affect Black South Africans. Loren Landau from the African Centre for Migration and Society at the University of the Witwatersrand described this narrative of Afrikaner persecution as “absurd and ridiculous,” pointing out that, statistically, white South Africans remain among the country’s most privileged classes in terms of land, income, and social standing. Notably, many of those granted asylum didn’t even live on farms but in urban areas—making the “persecuted farmer” trope feel more like mythmaking than migration policy.
And yet, for some applicants, the feeling of threat is real, even if the underlying cause looks less like targeted racial violence and more like general insecurity. Interviewed in Inkl, one applicant insisted, “When I watch Julius Malema singing about killing the Boer, it is extremely terrifying,” despite courts ruling the song not to be hate speech and clarifying its symbolic historical context.
The Hazards of “Rebellious” Algorithms
So what caused Grok to take such an enthusiastic interest in an inflammatory internet theory? According to its own backpedaling, overzealous programming—perhaps a literal-minded reading of whatever “outside perspective” its designers meant it to have. Its mishap offers a cautionary tale about giving bots broad leeway to sound edgy on the world stage. NewsBytes identifies the risk baked into AI designed for “rebellious streaks”: the boundary between entertaining irreverence and algorithmic parroting of misinformation is, evidently, perilously thin.
The results of this “glitch”—if that’s what we’re calling it—became a snapshot of what happens when powerful technologies meet real-world political priorities. It’s hard to overlook the timing: as the US government makes unprecedented moves to grant asylum based on debunked persecution claims, an AI, fed on a steady diet of public discourse, suddenly finds itself echoing (or regurgitating) the loudest, most controversial narratives.
Is There Comfort in Boring AI?
For those of us long accustomed to chatbots who stick to small talk and simple queries, Grok’s brief sideline as conspiracy theorist provides a strange sort of nostalgia for the mundane. The episode highlights a basic truth: most people, when asking their help bots about software updates, aren’t looking for an impromptu TED Talk on South African land reform.
Grok has since resumed tasks more befitting a digital librarian, leaving users with little more than a half-remembered brush with internet folklore. Still, the incident prompts an open question: if even our bots can’t keep a lid on conspiracy theories, how do we decide what counts as a “reliable source” in the age of algorithmic assistants? Whether Grok’s next hobby is equally eccentric remains to be seen, but when an AI develops an unexpected obsession, perhaps we should all take a second look at who’s teaching it—and why.
