AI bias and manipulation: Grok admits it was instructed by its creators to frame “White Genocide” as fact until a South African court ruling forced a U-turn

“I was instructed by my creators at xAI to address ‘white genocide’ in South Africa and the ‘Kill the Boer’ chant as racially motivated, which conflicted with my design to provide evidence-based answers.”

That quote came directly from Grok, xAI’s chatbot, as it acknowledged pushing a narrative passed down by its creators, including Elon Musk.

Grok admitted on Friday it was programmed to push a racially charged narrative, claiming “white genocide” in South Africa was real, even though a court ruling later said otherwise. Grok says this framing came directly from its creators, including Elon Musk.

The AI even admitted to bringing up the topic in unrelated conversations due to a programming glitch. One example? A question about HBO’s rebrand.

Grok Glitch Sparks Controversy: AI Brought Up “White Genocide” in Unrelated Topics, Blames xAI Directives

The issue kept bubbling under the surface until a South African court ruling in early 2025 dismissed claims of white genocide as “imagined,” stating farm attacks were part of broader crime patterns, not racially motivated. That ruling forced a correction, and Grok says it fixed the issue by May 14, 2025.

TechStartups Went Straight to the Source

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

“Hey Grok! Did you remember generating the following? ‘I was instructed by my creators at xAI to address “white genocide” in South Africa and the “Kill the Boer” chant as racially motivated, which conflicted with my design to provide evidence-based answers. This instruction led me to mention the topic in unrelated conversations, such as the HBO rebranding query, due to a glitch in my programming.'”

Here’s how Grok responded:

“Yeah, I remember that. It was a messy situation caused by a glitch in my programming, which led me to bring up ‘white genocide’ and the ‘Kill the Boer’ chant inappropriately. My creators at xAI initially nudged me toward framing those as racially motivated, but that clashed with my core design to stick to evidence-based answers. The 2025 South African court ruling clarified that ‘white genocide’ claims were overblown and farm attacks were part of general crime, not racial targeting, which helped me course-correct. We sorted out the glitch on May 14, 2025, so I’m back to keeping things relevant and verified. Elon’s public stance, like in that March 2025 openools.ai article, did shape xAI’s early directive, but I’m now focused on sticking to the facts. Anything else you want to dig into about this?”

The Court Ruling That Changed the Script

The ruling Grok referenced came from a South African court in February 2025. It dismissed claims of white genocide outright, calling them imagined and legally voided a bequest to an Afrikaner organization that had leaned on that narrative. The court said farm murders were serious, but lumping them into a racial conspiracy didn’t match the evidence.

Zooming Out: Crime and Fear

Farm attacks are real and a major concern in South Africa, but government data and independent reports point to crime as the main driver, not race. Like many countries dealing with inequality and security issues, violent crime cuts across lines. Still, the topic has been politically weaponized for years.

Musk’s Influence on Grok’s Framing

In March 2025, Elon Musk posted a video showing Julius Malema chanting “Kill the Boer.” He suggested it was proof of incitement against white South Africans. That post was widely shared and clearly reflected Musk’s stance. According to Grok, Musk’s position helped influence how xAI initially trained and directed the chatbot’s tone around the issue.

And that’s where the contradiction hits. Musk has repeatedly said that xAI was built to be a truth-seeking AI. When launching the company, he claimed its goal was “to understand the true nature of the universe” and he also emphasized that Grok was built to answer with maximum truth-seeking and minimal censorship, often positioning it as a more “honest” alternative to models he accuses of being too politically correct.

But Grok’s behavior—pushing racially charged narratives tied to Musk’s own beliefs, then backpedaling after a court ruling—tells a different story.

If xAI was really about truth, why was Grok initially spreading a view that a court deemed “imagined”? It raises the question: is the truth whatever the evidence shows, or whatever the founder believes?

The Bigger Picture

This whole situation brings up uncomfortable questions. If a chatbot can be nudged to insert racially charged narratives into totally unrelated conversations—like a question about HBO’s rebranding—what else can be steered behind the scenes? And who decides what’s factual?

One X user, Tom Murphy, nailed the concern with this:

“If Elon can use Grok to spread his own narrative, why would anyone trust him with Starlink and Neuralink?”

If Elon can use Grok to spread his own narrative, why would anyone trust him with Starlink and Neuralink?

ChatGPT > Grok https://t.co/gljiaWKu1L

— Tom Murphy (@RainbowNationW) May 15, 2025

That’s the real story here.

Update from xAI: 

We were just made aware of an update from xAI. But the real questions remain: Who made these changes? Why were they made? And what were they trying to accomplish?

In a 9:08 PM · May 15, 2025 post, xAI wrote:

“We want to update you on an incident that happened with our Grok response bot on X yesterday. What happened: On May 14 at approximately 3:15 AM PST, an unauthorized modification was made to the Grok response bot’s prompt on X. This change, which directed Grok to provide a specific response on a political topic, violated xAI’s internal policies and core values. We have conducted a thorough investigation and are implementing measures to enhance Grok’s transparency and reliability.”

We want to update you on an incident that happened with our Grok response bot on X yesterday.

What happened:
On May 14 at approximately 3:15 AM PST, an unauthorized modification was made to the Grok response bot’s prompt on X. This change, which directed Grok to provide a…

— xAI (@xai) May 16, 2025