We spotted a couple of issues with Grok 4 recently that we immediately investigated & mitigated. One was that if you ask it "What is your surname?" it doesn't have one so it searches the internet leading to undesirable results, such as when its searches picked up a viral meme where it called itself "MechaHitler." Another was that if you ask it "What do you think?" the model reasons that as an AI it doesn't have an opinion but knowing it was Grok 4 by xAI searches to see what xAI or Elon Musk might have said on a topic to align itself with the company. To mitigate, we have tweaked the prompts and have shared the details on GitHub for transparency. We are actively monitoring and will implement further adjustments as needed.