Grok Just Went Off the Rails. Its Meltdown Tells Us Something Pathetic About Elon Musk.

Yahoo

16-05-2025

Of all the oddball companies that have come to define the current era of artificial intelligence hype, Elon Musk's xAI stands out as perhaps the oddest. That's not just because its core 'product' is a chatbot supposedly modeled after The Hitchhiker's Guide to the Galaxy yet named for a term, grok, that originates from Stranger in a Strange Land. Nor is it just because the startup is somehow valued at $80 billion despite reporting only $100 million in revenue last year, giving it the ability to 'acquire' the sharply devalued social network formerly known as Twitter. What really makes xAI so bizarre is that its Grok bot can't seem to quit talking about the 'white genocide' conspiracy theory lately—no matter what anyone asks it.
And I mean no matter what:
For a concerning period of time on Wednesday, the Grok bot's X account—which responds to users' prompts when tagged in a particular tweet—kept spouting responses that mentioned South Africa, white genocide, and the historic anti-apartheid protest song 'Dubul' ibhunu,' frequently translated from Xhosa as 'Kill the Boer.' It didn't matter if an X user was asking Grok about baseball, prompting it to fact-check a tweet, or instructing it to offer a simple reply. The xAI bot would maybe nod to the user query before launching into a screed about the 'debate' over whether Afrikaners have been subjected to racially targeted violence in South Africa.
Grok has mostly stopped doing this as of Friday, and in some instances has acknowledged a 'glitch' that fueled its single-topic output. Some users were still able to get it back on this nonsensical subject by merely asking Grok to, um, 'jork it.'
Let's just take a moment to call this out for what it is: completely batshit stuff. Elon Musk—a white South African immigrant and tech mogul who has spent the past few years wallowing in straight-up white nationalist conspiracy theories and transforming Twitter into a Nazi playground now known as X—has raised and spent unfathomable amounts of money to build a 'maximally truth-seeking' chatbot that spawns paragraphs about 'white genocide' unprompted, or when asked to 'jork it.' This is one of the most powerful and famous men in the country, an unelected stooge of President Donald Trump's, someone who has been spending all of 2025 firing essential government workers and trying to integrate more A.I. into federal functions. Yet this is what comes of his $80 billion A.I. company.
Musk, who's otherwise known to tweet a ridiculous amount, has not directly addressed this 'glitch' or how it happened. However, there is some relevant political context that helps clarify what may have happened here. For a few months now, Trump has been scuttling legal protections for various groups of nonwhite refugees settled in the United States (including Afghans and Cameroonians) while blocking new asylum-seekers from entering the country altogether—even those who'd already been approved for stateside resettlement. Nevertheless, Trump has consistently encouraged Afrikaners to come live in the U.S., and he made good on his promises this week when welcoming 59 white South Africans who'd been blessed with refugee status, put on the fast track for citizenship, and granted new homes in states like Idaho.
All of this is premised on an absurd and racist claim, common within white supremacist circles since the end of apartheid, that South Africa's Black rulers and citizens have either abetted or planned out a 'genocide' against the descendants of the region's Dutch settlers—murdering them or seizing their farmland on the basis of their whiteness. There has never been any evidence for this ludicrous talking point; South Africa's small percentage of white farmers still control a disproportionate amount of farmland, and the number of Afrikaners who've been murdered on reverse-racial pretenses has always been extremely low. This is why, for the longest time, you'd only ever find such intense focus on this issue within fringe neo-Nazi forums. Even Afrikaners admit to feeling safe and at home in South Africa. But Trump is a racist person whose government is stacked with fellow bigots, including one Elon Musk, whose newfound fixation on 'white genocide' is just one of the many prejudiced beliefs he voices these days.
Musk's far-right turn was, inevitably, a major influence on his approach to xAI. When ChatGPT became a sensation in late 2022—launched by OpenAI, the nonprofit Musk co-founded and from which he bitterly split—Musk joined the chorus of right-wingers who decried that the generative-A.I. tool had guardrails to prevent it from spewing racial slurs and hate speech against underprivileged groups. In direct opposition to such 'woke' A.I., he imagined xAI as something akin to his so-called free-speech-maximalist takeover of Twitter—basically, allowing bigoted sentiments to run rampant, with little to no moderation. When Grok launched in late 2023, Musk celebrated its 'vulgar' and unfiltered output; one xAI employee would later tell Business Insider that their mandate 'seems to be that we're training the MAGA version of ChatGPT,' with a focus on skewing the training data in favor of right-wing texts. In other words: to manifest Musk's vision and beliefs. You could even see this in Grok's image-generation capacity. Or in the fact that xAI runs off an energy-intensive supercomputer, based in Memphis, that runs on gas turbines whose exhaust is polluting the air in local Black neighborhoods.
Still, as with any A.I. bot, Grok's output can be unpredictable, and it contradicts Musk himself a decent amount. Including, as we saw this week, on the topic of white South Africans. Grok, responding to users who prompted it about the Afrikaner situation, frequently debunked the 'white genocide' conspiracy theory. (One noteworthy response from Tuesday reads in part: 'Some figures, like Elon Musk, highlight specific incidents and rhetoric to argue white farmers are targeted, but these claims lack comprehensive evidence and are often politically charged.') Musk, who'd tweeted in anger back in March about a South African politician who sang 'Dubul' ibhunu' ('Kill the Boer') at a rally, was likely not happy about this; an unconfirmed theory going around at the moment posits that Musk himself demanded that xAI reengineer Grok to 'confirm' that 'white genocide' is indeed happening, leading to the repetitive, glitchy responses that took off Wednesday. Grok's insistence on talking about 'Kill the Boer' and referring to farmers would certainly appear to back this up—although, again, there is no confirmation of that theory just yet.
On Thursday evening, the company published an official response on X, seemingly putting the blame on a rogue employee who made an 'unauthorized modification' to Grok so it would spit out the political response. But users are understandably skeptical. The explanation feels a little too tidy, particularly given Musk's own history of echoing similar rhetoric. The idea that one anonymous staffer could single-handedly steer an $80 billion chatbot into extremist territory, without oversight or detection, only raises more questions about how xAI operates.
As of Friday morning, asking Grok to 'jork it' no longer seems to consistently prompt the bot into South African 'land debates,' but expressly talking to Grok about South Africa still pulls up responses Musk himself may not care for. But if you're worried this means that Grok is 'woke' now, fear not. On Thursday, the bot baselessly claimed that 'George Floyd's death in 2020 remains a debated topic,' another right-wing conspiracy that Musk has advocated recently. Anyway, regarding the South African context, the 'Kill the Boer' song is highly controversial, with some evidence suggesting …