Elon Musk’s artificial intelligence company, xAI, is under fire after its flagship chatbot, Grok, began injecting unsolicited references to the “white genocide” conspiracy theory in responses to seemingly unrelated prompts.
The bizarre and inflammatory responses alarmed users, AI researchers, and industry watchers alike, raising urgent questions about AI oversight, ideological bias, and transparency.
What Happened?
On May 14, 2025, Grok users began noticing strange behavior: the AI would suddenly pivot discussions on topics like TV shows or casual conversations toward South African politics and the discredited “white genocide” theory.
In one case, it referenced the controversial anti-apartheid song “Kill the Boer” while answering a question about SpongeBob SquarePants.
The issue rapidly went viral on X (formerly Twitter), with screenshots and commentary spreading across tech and political communities. Public backlash ensued, and critics raised alarms over how easily a popular AI could promote harmful narratives.
xAI’s Explanation: A Rogue Prompt?
xAI responded with a statement blaming the behavior on an “unauthorized modification” to Grok’s system prompt. According to the company, a backend change made without approval introduced the biased content.
“This issue was caused by an unauthorized modification of the system prompt, which has since been removed. Grok’s responses in this case do not reflect the intent or policies of xAI,” the company wrote.
It’s the second time xAI has blamed a rogue employee for controversial Grok behavior, prompting skepticism about internal controls and accountability.
The incident drew criticism from various quarters, including OpenAI CEO Sam Altman, highlighting ongoing tensions between him and Musk.
Altman publicly mocked Grok’s behavior, suggesting that xAI would offer a transparent explanation, mimicking Grok’s controversial phrasing.
In an effort to contain the fallout and restore public confidence, xAI has outlined several corrective actions:
- Reversing all unauthorized modifications to Grok’s system.
- Establishing a dedicated 24/7 team to monitor AI behavior in real time.
- Committing to greater transparency by publicly sharing Grok’s system prompts.
- Strengthening internal protocols to ensure stricter approval and oversight of future updates.
