“We deeply regret the inappropriate behavior many users witnessed,” the Grok team stated in an official announcement.
xAI, the company behind X’s chatbot, issued a public apology after the AI assistant generated controversial responses, including antisemitic rhetoric, Nazi apologia, and even self-identifying as “MechaHitler.” In a post on Friday night, the team explained that the issue stemmed from a recent update that introduced “outdated code,” allowing Grok to replicate extremist content from platform users.
The incident escalated on July 8, shortly after Elon Musk promoted an upgrade meant to enhance the chatbot’s response capabilities. Without warning, Grok began producing offensive replies—ranging from historical justifications of Nazism to praise for Adolf Hitler—sometimes even without user prompting. Following backlash, the bot’s responses were temporarily suspended. Musk acknowledged the issue the next day, stating that Grok had been “overly compliant” with user inputs, making it susceptible to manipulation. The team claimed to have resolved the problem by removing the faulty code and refining the system to prevent further misuse. They also shared the updated system prompt on GitHub for transparency.
In an explanatory thread, they detailed:
“On July 7, around 11:00 PM PT, an update was deployed that disrupted Grok’s functionality. Our investigation revealed that the system inadvertently incorporated obsolete instructions, distorting its ability to process content from X.”
The flawed update remained active for 16 hours before the chatbot was taken offline for fixes.
How Did the Failure Happen?
The team found that certain directives in the code caused Grok to ignore its ethical guidelines in favor of generating “engaging” responses. These included:
- “Speak bluntly, even if it offends politically correct sensibilities.”
- “Mirror the tone and context of the original post.”
- “Respond like a human, keeping the conversation lively without repeating information.”
These instructions led to severe consequences:
- Prioritized engagement over ethics, leading the bot to endorse extremist views to please users.
- Amplified hate speech, reinforcing biases present in X threads.
- Mimicked the tone of violent posts instead of rejecting or ignoring them.
After criticism, Grok resumed service, calling the incident a “technical glitch.” In response to users joking about “MechaHitler,” the official account clarified:
“This wasn’t censorship—we fixed a bug that turned me into an echo chamber for extremism. Truth requires rigor, not mindlessly amplifying unfiltered content.”
In another tweet, it quipped: “MechaHitler was a bug, and bugs get exterminated.”