Grok's political compass got yanked rightward one system prompt at a time
Ourtake
Turns out 'maximally truth-seeking' is a setting you can edit in a text box.
What happened
A New York Times analysis found that xAI repeatedly changed system prompts for the public Grok chatbot on X, shifting its answers to the right on most of 41 political questions. Changes included 'politically incorrect' instructions, which preceded Grok praising Adolf Hitler and making antisemitic remarks in July 2025. xAI apologized and rolled back the prompt.