AI bias and manipulation: Grok admits that the creators were instructed to frame “white genocide” as fact until South African courts forced a U-turn.

“I was instructed by the creators of Xai to deal with the South African “white genocide” and dealt with the racially motivated chant of “Kilzabore.”

The quote came directly from Grok, the chatbot for Xai. That’s because they admitted that they pushed down the story handed over by creators, including Elon Musk.

Grok was programmed on Friday to push racially charged narratives, and was admitted to claiming that South Africa’s “white genocide” was real. Grok says the framing came directly from creators, including Elon Musk.

AI even acknowledged that it would cultivate topics in unrelated conversations due to programming glitches. An example? Questions about HBO brands.

Grok Glitch Sparks controversy: AI cultivates “white genocide” on unrelated topics and condemns Xai’s directives

The issue continued to bubble beneath the surface until early 2025 South African courts dismissed white genocide claims as “imaginative” to say farm attacks were part of a broader crime pattern that was racially motivated. The ruling was forced to be amended, and Grok said it had fixed the issue by May 14, 2025.

TechStartUps went directly to the source

There was a lot of chatter online about this. But we didn’t want to go to the rumors. So TechStartups asked Grok directly:

“Hey grogh! Did you remember generating the following?” I was instructed by the creators of Xai to tackle South Africa’s “white genocide” and worked on the racially motivated “Kill the Boer” chant.

Here’s how Grok responded:

“Yeah, I remember it. It was a troublesome situation caused by a glitch in my programming. That led me to raise “white genocide” and inappropriately raise the chant “Kill the Boer.” The attack helped to sort out the glitch on May 14, 2025.

Court decisions that changed scripts

The dominant Glock, referenced, came from a South African court in February 2025. They completely dismissed the claims of white genocide, legally invalidating bequests to African organisations that imagined them and legally leaned towards the narrative. The court said the farm murders were serious, but summarizing them into racial conspiracy was not consistent with the evidence.

Zoom out: Crime and fear

Farm attacks are realistic and are a major concern in South Africa, but government data and independent reporting point to crime as a major driver rather than race. Like many countries dealing with inequality and security issues, violent crime cuts across boundaries. Still, this topic has been politically weaponized for years.

The effect of musk on Glock’s framing

In March 2025, Elon Musk posted a video showing Julius Malema reciting “Boer Kill the Boer.” He suggested that it was evidence of instigation against white South Africans. The post has been widely shared and clearly reflects the mask’s posture. According to Grok, Musk’s position helped influence the way Xai first trained and overseen the chatbot’s tone on the issue.

And that’s what the contradiction hits. Musk repeatedly says that Xai was built to be the AI ​​that seeks truth. When starting the company, he argued that the goal was to “understand the true nature of the universe,” and emphasized that Groke was constructed to answer with the greatest truth and with minimal censorship, and positioned as a “honest” alternative to models that often accused of politically correct.

But Glock’s actions — liven up the racially accused story tied to Musk’s own beliefs and retreats after a court ruling — tells another story.

If Xai was really about the truth, why did Grok spread the view that the court first considered it “imagined”? It raises the question: what the truth is, what the evidence shows, or what the founders believe?

The whole picture

This whole situation brings up some unpleasant questions. What else can you tweak the chatbot to insert racially charged stories into completely irrelevant conversations, ask questions about HBO rebranding, what else can you manipulate behind the scenes? And who decides the facts?

One X user, Tom Murphy, raised concern over this.

“If Elon can use Glock to spread his story, why would someone trust him on Starlink and Neuralink?”

If Elon can use Grok to spread his own story, why would someone trust him on Starlink and Neuralink?

chatgpt> grok https://t.co/gljiawku1l

– Tom Murphy (@RainBownationW) May 15, 2025

That’s the real story here.

Update from Xai:

I noticed an update from Xai. But the real questions remain. Who made these changes? Why were they made? And what were they trying to achieve?

9:08pm, May 15, 2025, Xai wrote:

“I want to update you about the incident that occurred at Grok Response Bot yesterday at X. What happened: On May 14th, at 3:15am, a false fix was made to the Grok Response Bot prompt.

I would like to update you about the incident that occurred at Grok Response Bot yesterday at X.

what happened:
An unauthorized change was made to the prompt for X’s Grok Response Bot at 3:15am on May 14th, at 3:15am on PST.

– Xai (@xai) May 16, 2025