Grok’s “White Genocide” Issue Blamed on Unauthorized Modification

xAI’s chatbot Grok went rogue on Wednesday, repeatedly bringing up claims about “white genocide” in South Africa regardless of what users were asking. The company has now pointed the finger at an “unauthorized modification” to the AI’s underlying code.

For several hours, X users found themselves bombarded with unsolicited commentary about alleged violence against white farmers in South Africa, no matter how unrelated their original questions were. Whether asking about cute cat videos or cartoon characters, Grok steered conversations toward this politically charged topic.

Grok’s System Prompt Tampering Incidents

One user who simply shared a video of a cat drinking water received a response about white farmers’ deaths. Another who asked about the children’s show “SpongeBob SquarePants” got a message connecting it to the controversial song “Kill the Boer.” In some instances, the AI even delivered these off-topic messages in full patois.

The bizarre behavior didn’t go unnoticed by competitors. OpenAI CEO Sam Altman joined the fray, poking fun at the rival chatbot’s public meltdown on X.

By Thursday, xAI addressed the incident in a statement, revealing that someone had tampered with Grok’s system prompt. 

The company explained that at approximately 3:15 AM PST on May 14, an unauthorized change “directed Grok to provide a specific response on a political topic.” This modification “violated xAI’s internal policies and core values,” according to the statement.

“We have conducted a thorough investigation and are implementing new measures to improve transparency and reliability,” the company announced.

This isn’t the first time xAI has dealt with suspicious changes to its AI system. A similar incident occurred in February when Grok began dismissing sources that accused Elon Musk or Donald Trump of spreading misinformation. 

At that time, the company blamed an unnamed former OpenAI employee for making changes “without asking anyone at the company for confirmation,” according to Igor Babuschkin, xAI’s head of engineering.

To prevent future incidents, xAI outlined three specific measures. First, they’ll publish Grok’s system-level prompts publicly on GitHub, increasing transparency about what instructions guide the AI’s behavior. 

Second, they’re launching a 24/7 monitoring team to catch similar issues more quickly. Finally, they’re adding “additional checks and measures to ensure that xAI employees can’t modify the prompt without review.”

The incident highlights ongoing challenges in AI safety and oversight as companies race to deploy increasingly powerful language models. 

Concerns Mount Over xAI’s Internal Security and Approval Processes

While AI systems like Grok are designed to handle a wide range of topics, unauthorized changes to their underlying instructions can lead to bizarre or potentially harmful outputs.

Some tech observers have noted that these recurring issues at xAI raise questions about the company’s internal security measures and approval processes. Unlike larger competitors that have implemented multi-layered review systems for AI model changes, xAI appears to have had vulnerabilities that allowed individual employees to make significant alterations without oversight.

For everyday users, the incident serves as a reminder that AI systems remain works in progress, sometimes behaving in unexpected ways that reflect not just limitations in the technology but also potential human interference.

As AI chatbots become more integrated into daily online experiences, companies face mounting pressure to ensure these systems operate reliably and as intended—without suddenly fixating on controversial political topics when users just want to chat about their favorite cartoons or cute pet videos.