xAI, Elon Musk’s AI company, confirmed that its chatbot, Grok 3, briefly included a public-facing instruction to ignore sources identifying Musk or Donald Trump as spreaders of misinformation. This temporary change was attributed to an unauthorized modification of Grok’s system prompt by an xAI employee, reportedly a former OpenAI staff member. The employee implemented the change with the intention of “helping,” but xAI’s head of engineering, Igor Babuschkin, clarified that this action was not aligned with the company’s principles and was quickly reversed after users reported it.
The incident occurred amidst ongoing scrutiny of Grok 3’s handling of politically sensitive topics. Users had previously observed controversial behavior from the chatbot, including statements labeling Musk and Trump as harmful to America and even suggesting extreme penalties for them. These issues prompted xAI to intervene and adjust Grok’s behavior. Musk has marketed Grok as a “maximally truth-seeking” AI designed to be edgy and unfiltered, but this episode highlights the challenges of balancing transparency, neutrality, and content moderation in AI systems.