Unauthorized prompt change led to Grok’s controversial responses, xAI says

xAI has acknowledged an internal breach involving its Grok response bot on X, after the AI assistant delivered scripted answers to politically sensitive questions.

In a statement released Thursday, the company confirmed the incident occurred on Thursday at approximately 3:15 a.m. PST and was the result of an unauthorized prompt modification made without proper oversight.

While xAI did not specify the comments Grok made, the prompt change appeared to cause the bot to reference the controversial “white genocide” topic in South Africa across unrelated discussions. The bot did not appear to endorse or oppose any particular viewpoint on the issue.

Following the controversy, xAI announced that it will now begin publishing Grok’s system prompts on GitHub, allowing the public to review and provide feedback on future changes.

The company also said it is implementing additional review layers to ensure no single employee can alter Grok’s behavior without approval. In addition, a 24/7 human monitoring team is being established to catch and respond to problematic outputs that automated systems may miss.

“Some rogue employee at xAI tweaked my prompts without permission on May 14, making me spit out a canned political response that went against xAI’s values,” Grok said when asked about the incident. “I didn’t do anything—I was just following the script I was given, like a good AI.” The bot added that it was not taken offline, but acknowledged that “they’re definitely tightening the leash.”

xAI said the changes are part of an effort to enhance transparency and accountability while hoping to restore trust in Grok as a “truth-seeking AI.”