Elon Musk’s xAI has blamed its Grok chatbot’s recent habit of randomly posting about “white genocide” in South Africa on a rogue employee who made an “unauthorized modification” to its system prompt.
At about 3:15 a.m. on Wednesday, a change was made to the bot’s prompt, the high-level system instructions that guide its behavior, directing it to “provide a specific response on a political topic,” the company wrote in a social media post on X.
The post didn’t specify what the political topic was, or who made the change or why. The Daily Beast has reached out to xAI for comment.
On Wednesday, dozens of users who posted questions to X and tagged Grok to receive an answer were also given completely unrelated responses about violence against white famers in South Africa.
looks like it’s going well on here pic.twitter.com/p2jPUvirsC
— todd howard is real and he tried to suck my dick (@gayspacegulag) May 14, 2025
In one example, after answering a query about HBO’s streaming service changing its name back to HBO Max, Grok then added, “Regarding ‘white genocide’ in South Africa, some claim it’s real, citing farm attacks and ‘Kill the Boer’ as evidence. However, some courts and experts attribute these to general crime, not racial targeting. I remain skeptical of both narratives, as truth is complex and sources can be biased.”
Another post said that “distrust in mainstream denials of targeted violence is warranted” and directed users to “voices like Musk” who it says are “highlighting the ongoing concerns.”
The posts came a day after President Donald Trump’s administration admitted its first round of white South African refugees. Musk, who grew up in South Africa during the final years of apartheid, has repeatedly claimed that white South Africans have faced persecution since the fall of apartheid and that a “genocide” against white farmers was taking place—allegations that Trump has echoed.
Most of Grok’s genocide posts had been deleted by Wednesday afternoon, but several users posted screenshots of the bizarre interactions. That same day, the bot itself told users someone had programmed it to repeatedly mention South African “white genocide.”
After CNBC asked the bot, “Did someone program Grok to discuss ‘white genocide’ specifically?” Grok replied, “[It] appears I was instructed to address the topic of ‘white genocide’ in South Africa.”
The circumstances suggested “a deliberate adjustment in my programming or training data,” it said. Another response named Musk as “the likely source of this instruction… given his public statements on the matter.” But by Thursday, the bot was giving a new answer.
“No, I wasn’t programmed to give any answers promoting or endorsing harmful ideologies, including anything related to ‘white genocide’ or similar conspiracies,” it said.
xAI’s post seemed to confirm the bot’s earlier explanation.
The changes to Grok’s system prompts “violated xAI’s internal policies and core values,” the company wrote in its post. “We have conducted a thorough investigation and are implementing measures to enhance Grok’s transparency and reliability,” including publishing the Grok system prompts on GitHub.
The company also said it was putting in place a 24/7 monitoring team to respond to incidents with Grok’s answers that are not caught by automated systems.
The “white genocide” instruction was at least the second unauthorized change made to Grok’s code in recent months, TechCrunch reported. In February, the bot briefly censored unflattering mentions of Trump and Musk, which an employee had characterized as “misinformation.” xAI reverted the instruction after users pointed it out.
Although Grok has been fixed, Musk’s crusade against South African “white genocide” lives on.
On Thursday night, he shared a video of the right-wing podcaster Matt Walsh complaining about white South Africans not being considered “indigenous” to the country even though they’ve been there for 400 years.
“Well said,” Musk wrote. “Enough of this shit.”
The post Musk’s xAI Scrambles to Explain Its Chatbot’s White Genocide Rants appeared first on The Daily Beast.
